The Giant Cretaceous Coelacanth (Actinistia, Sarcopterygii) Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, and Its Bearing on Latimerioidei Interrelationships

Hugo Dutel1,2*, John G. Maisey3, David R. Schwimmer4, Philippe Janvier1, Marc Herbin2, Gae¨l Cle´ment3 1 UMR 7207 CNRS-MNHN, Centre de Recherches sur la Pale´obiodiversite´ et les Pale´oenvironnements (CR2P), De´partement Histoire de la Terre, Muse´um national d’Histoire naturelle, Paris, France, 2 UMR 7179 CNRS-MNHN, Me´canismes adaptatifs: des Organismes aux Communaute´s, De´partement E´cologie et Gestion de la Biodiversite´, Muse´um national d’Histoire naturelle, Paris, France, 3 Division of Paleontology, American Museum of Natural History, City of New York, New York, United States of America, 4 Department of Earth and Space Sciences, Columbus State University, Columbus, Georgia, United States of America

Abstract We present a redescription of Megalocoelacanthus dobiei, a giant fossil coelacanth from Upper Cretaceous strata of North America. Megalocoelacanthus has been previously described on the basis of composite material that consisted of isolated elements. Consequently, many aspects of its anatomy have remained unknown as well as its phylogenetic relationships with other coelacanths. Previous studies have suggested that Megalocoelacanthus is closer to Latimeria and Macropoma than to Mawsonia. However, this assumption was based only on the overall similarity of few anatomical features, rather than on a phylogenetic character analysis. A new, and outstandingly preserved specimen from the Niobrara Formation in Kansas allows the detailed description of the skull of Megalocoelacanthus and elucidation of its phylogenetic relationships with other coelacanths. Although strongly flattened, the skull and jaws are well preserved and show many derived features that are shared with Latimeriidae such as Latimeria, Macropoma and Libys. Notably, the parietonasal shield is narrow and flanked by very large, continuous vacuities forming the supraorbital sensory line canal. Such an unusual morphology is also known in Libys. Some other features of Megalocoelacanthus, such as its large size and the absence of teeth are shared with the mawsoniid genera Mawsonia and Axelrodichthys. Our cladistic analysis supports the sister-group relationship of Megalocoelacanthus and Libys within Latimeriidae. This topology suggests that toothless, large-sized coelacanths evolved independently in both Latimeriidae and Mawsoniidae during the Mesozoic. Based on previous topologies and on ours, we then review the high-level taxonomy of Latimerioidei and propose new systematic phylogenetic definitions.

Citation: Dutel H, Maisey JG, Schwimmer DR, Janvier P, Herbin M, et al. (2012) The Giant Cretaceous Coelacanth (Actinistia, Sarcopterygii) Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, and Its Bearing on Latimerioidei Interrelationships. PLoS ONE 7(11): e49911. doi:10.1371/journal.pone.0049911 Editor: Daphne Soares, University of Maryland, United States of America Received August 7, 2012; Accepted October 15, 2012; Published November 27, 2012 This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Funding: GC and DS were supported by the American Museum of Natural History (AMNH) Axelrod Fund for his visit in the AMNH fossil fish collection to work on the herein described material. This work was partly supported by the French Agence Nationale de la Recherche under the TERRES Project (ANR-2010-BLAN-607- 03). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. * E-mail: [email protected]

Introduction features evolved independently in Mawsonia and Megalocoelacanthus. The authors have suggested that Megalocoelacanthus is closer to Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994 is Macropoma and Latimeria than to Mawsonia. However, their a giant marine coelacanth discovered in 1987 [1] in the Upper assumption was only based on the comparison of meristic data Cretaceous Bluffown Formation, southeastern USA. With an of some anatomical features (table 2 in [1]), but not on a estimated length of 3.5 m, Megalocoelacanthus is among the largest phylogenetic analysis. known coelacanths. Similar dimensions (i.e. more than 2.0 m in Since its discovery, the relationships of Megalocoelacanthus with total length) are reached by the late Jurassic-mid Cretaceous genus other coelacanths have never been investigated. Here we present a Mawsonia from North Africa and South America [2–7], and the description of cranial and postcranial skeleton of Megalocoelacanthus Early Cretaceous genus Axelrodichthys from Brazil [5] also attained based on new and holotype material. A phylogenetic analysis of 39 a large size. Previous phylogenetic analyses [8–14] supported a taxa and 110 characters is performed to clarify the position of sister-group relationship of Mawsonia and Axelrodichthys within Megalocoelacanthus among coelacanths, and its bearings on Meso- Mawsoniidae Schultze, 1993 [15]. This close affinity between zoic coelacanth’s interrelationships are subsequently discussed. toothless, large-sized Mesozoic coelacanths raised the question Implications for the coelacanth taxonomy will also be discussed whether these features could be synapomorphies of a putative and the application of phylogenetic definitions to coelacanth clade or have evolved independently in several lineages. The taxonomy will be proposed based on our novel topology. evolution of large sized, toothless coelacanths has been briefly discussed by Schwimmer et al. [1] who suggested that these

PLOS ONE | www.plosone.org 1 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Materials and Methods ethmosphenoid and otoccipital portions), snout, lower jaws, gular plates, branchial arches, urohyal, hyoid skeleton, and shoulder 1. Geological context girdle. Although most of these isolated remains are strongly Megalocoelacanthus remains are known only from the United flattened laterally, they are outstandingly preserved. Significant States. The holotype CCK 88-2-1 consists of cranial and elements from the holotype specimen CCK 88-2-1 are also branchials elements [1], and was found in the lower part of the included in the description for further comments. Blufftown Formation in eastern Alabama, that is early Campanian in age (Figure 1). Remains of Megalocoelacanthus were also first 3. Nomenclatural Acts reported in five additional regional localities from the southeastern The electronic version of this document does not represent a states of Alabama and Georgia [1], along with a single coronoid published work according to the International Code of Zoological fragment from New Jersey. Subsequently, unpublished specimens Nomenclature (ICZN), and hence the nomenclatural acts have been collected in coeval strata in Mississippi (Schwimmer, contained in the electronic version are not available under that Earl Manning, pers. comm.), and Kansas. All known fossils of Code from the electronic edition. Therefore, a separate edition of Megalocoelacanthus with well-known stratigraphic associations are of this document was produced by a method that assures numerous late Santonian to mid-Campanian age [1], except the New Jersey identical and durable copies, and those copies were simultaneously specimen which is in a late Campanian-early Maastrichtian obtainable (from the publication date noted on the first page of this deposit. However, the latter fossil is a highly ablated principal article) for the purpose of providing a public and permanent coronoid fragment, preserved in a near shore lag deposit, which scientific record, in accordance with Article 8.1 of the Code. The may have been reworked from older material: its age is thus separate print-only edition is available on request from PLoS by uncertain. sending a request to PLoS ONE, Public Library of Science, 1160 Battery Street, Suite 100, San Francisco, CA 94111, USA along 2. Material with a check for $10 (to cover printing and postage) payable to This redescription of Megalocoelacanthus is mainly based on ‘‘Public Library of Science’’. AMNH FF 20267, which was collected in 2007 in the Niobrara In addition, this published work and the nomenclatural acts it Formation, in the Northern Lane County, Kansas (Figure 1). It is contains have been registered in ZooBank, the proposed online early Campanian in age and thus approximately coeval with the registration system for the ICZN. The ZooBank LSIDs (Life holotype CCK 88-2-1. The new specimen consists of skull (both Science Identifiers) can be resolved and the associated information

Figure 1. Geological context of Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994. Left, white star indicates the geographic location of the locality of AMNH FF 20267 in the Niobrara chalk of Lane County Kansas, USA (modified from http://www.kgs.ku.edu/ General/Geology/County/klm/lane.html). Right, stratigraphic correlation chart between the principal stratigraphic units from North America (taken from [1]). The Niobrara Formation in Lane County, Kansas, is correlated with the Blufftown, Mooreville, and Eutaw Formations in Alabama and Georgia, where the first occurrences of Megalocoelacanthus were reported. doi:10.1371/journal.pone.0049911.g001

PLOS ONE | www.plosone.org 2 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus viewed through any standard web browser by appending the LSID thus reflect the actual width of the skull roof as in Whiteia [12] and to the prefix ‘‘http://zoobank.org/’’. The LSID for this Macropoma (figures 3.15, 3.19A in [12]). The tip of the snout publication is: urn:lsid:zoobank.org:pub:6F452EF9-02A6-4B3F- appears to be strongly consolidated due to the tight suture between 9E0B-3E4C14BADC80. the premaxilla and the median rostral (Figure 4A). Although the snout is heavily ossified, it may have been loosely attached to the Results lateral ethmoid, anterior nasal, and tectal (if present). This condition is also observed in Macropoma lewesiensis (BMNH 4207). 1. Anatomical description The entire surface of the snout is ornamented with coarse 1.1 Dermal bones of the skull roof. Despite strong lateral rugosities, making it difficult to observe the suture between the compression, all elements are very well preserved and allow bones precisely. The premaxilla is robust, bears no teeth, and is detailed description. The parietonasal shield (Figures 2, 3, 4) is separated at the symphysis from its antimere by a large longer than the postparietal shield (Figures 5, 6, 7, 8) very narrow trapezoidal, median pore of the sensory line canal (m.p.s.c, and straight in lateral view. The intracranial joint is transversely Figure 4A). Such a condition is also observed in Macropoma straight. The parietonasal series consists of two pairs of parietals (figure 3.20B in [12]) and Latimeria (figure 8 p.s.m.e.m in [18]), but (Pa.a, Pa.p, Figures 2, 3), but the number of nasals cannot be the median pore of the sensory line canal is much larger in assessed, and only the posteriormost nasal (Na, Figure 2) can be Megalocoelacanthus. The ventral half of the premaxilla is very broad observed. However, its relationships with the elements of the and the base of the bone extends medially as a thin flattened anteriormost part of the ethmosphenoid portion of the skull are surface that closes ventrally the median pore of the sensory line very difficult to observe. The bones of the parietonasal series are canal (Figure 4A). The base of the premaxilla meets its antimere in slender and elongated. The posterior parietals are the largest of the a recess, posterior to the opening of the median pore of the sensory series as in Macropoma, Holophagus [12] and Swenzia [8,16], and line canal. The anteroventral margin of the premaxilla is slightly their center of ossification is situated posteriorly. The posterior folded ventrally, and crenate. Ventrally, the right premaxilla parietals meet the anterior parietals in a transverse indented presents a raised area pierced by a pore that is directed suture. The median suture between the two paired parietals is posteroventrally. The dorsal lamina of the premaxilla (d.l.Pmx, straight and only leaves a narrow gap on the midline, probably due Figure 4A) is well expanded and forms the lateral margin of the to the settling of the bones under lateral constraints during opening of the anterior tube of the rostral organ (a.ros, Figure 4A). fossilization. The sutures between the different parietonasal The median rostral (ros.m, Figure 4A, 4B), usually poorly elements (posterior parietal/anterior parietal; anterior parietal/ preserved in coelacanths, is here complete and in its natural nasal) extend far laterally, and extend into the vacuity of the position. It is cross-shaped, shorter than broad, and shows no supraorbital sensory line canal at the level of the contact between indication of strong distortion. It is raised anteroposteriorly, and its the dermal bones (Figure 2). The supraorbital series can be surface is ornamented with coarse rugosities. On both side, it distinguished by the presence of sutures between to the shows a curved, posteroventral extension towards the dorsal parietonasal series and the pillars that are crossing the supraorbital lamina of the premaxilla. The better-preserved right side of the sensory line canal (So, pi, so.s.c, Figures 2, 3). However their snout allows the description of the relations between these bones precise number cannot be assessed due to the poor preservation of (Figure 4A). The median rostral reaches the lateral tectal medial to this area of the skull roof. On both side of the skull, an elbowed the premaxilla, and forms the dorsal margin of the opening of the process (?v.pr.Pa, Figures 2, 3) extends just below the margin of anterior tube of the rostral organ. Posteriorly, the median rostral joint between the posterior pair of parietal and the postparietal. caps the anterior wall of a large and deep ovoid cavity (n.c, This process could correspond to the pedicel of the ventral process Figure 4B). The lateral wall, formed anteriorly by the lateral of the parietal, which forms a ridge that slides in the groove on the extension of the median rostral, is then extended by the lateral postparietal. rostral that curves medially in its posterior portion. The ovoid The lateral rostral (L.r, Figures 2, 3) is a large bone which cavity is paired, but not separated from its antimere by a medial extends posteroventrally as a relatively short and flattened tube septum. Indeed, it is clearly individualized on either side of the that encloses the infraorbital sensory line canal. There is no snout by a medial, bell-shaped cavity (Figure 4B). It is unlikely that evidence of grooves on its dorsal surface, unlike in Diplurus, Macropoma, Latimeria, Laugia [12], Swenzia [8], and Rhabdoderma [17]. this paired cavity corresponds to the rostral organ cavity of The lateral rostral usually separates the anterior nostril from the Latimeria, because the latter is median, and usually situated more posterior nostril. Anteriorly, the ventral process of the lateral posteriorly along the body axis, posterodorsally to the nasal rostral (v.pr.L.r, Figures 3) is sutured to the lateral ethmoid, and its capsule. Consequently, the anterior portion of the rostral organ anterior margin is notched and corresponds to the posterior lies dorsal to the olfactory capsules and the internasal septum. The margin of the opening for the anterior nostril (nos.a, Figure 3) as in very anterior position, the individualization of the lateral cavities, Macropoma (figure 6.10 in [12]). On both sides, the posterior and their openings towards the exterior, suggests that they housed margin of the lateral rostral is notched dorsally to the tube the nasal capsules. As in Latimeria, the cavities are oriented enclosing the infraorbital sensory line canal. A space is clearly anteromedially. The right nasal cavity is opened by a canal observable on the left side, between the lateral rostral and the directed laterally, that pierces the lateral rostral (c.nos.a, Figure 4B) supraorbital series and suggests that the position of the opening of and opens ventrally to the notch formed by the contact of the the posterior nostril (nos.p, Figure 3) was similar to that in other dorsal lamina of the premaxilla and the lateral rostral. This coelacanths. On the left side, a conspicuous suture is observable opening is interpreted as being the anterior nostril (nos.a, between the lateral rostral and the supraorbital series. Figure 4A). Another foramen pierces the anterior wall of the The tip of the snout is preserved in three dimensions, but is nasal capsule towards the opening of the anterior tube of the isolated from the rest of the skull (Figure 4). The snout is partially rostral organ (f.a.n.c, Figure 4A, 4B). The medial cavity that fused and consists of a pair of premaxilla, a median rostral, and the separates the nasal capsules is partly closed ventrally by the right anterior portion of the lateral rostral (Pmx, ros.m, L.r, flattened base of the premaxillaries, and opens anteriorly through Figure 4A). It shows no trace of strong distortion: its shape may the median pore of the sensory line canal.

PLOS ONE | www.plosone.org 3 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 2. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Ethmosphenoid portion of the skull in right lateral view. Abbreviations: ant.com.so.s.c, anterior commissure of supraorbital sensory line canal; ant.pr, antotic process; a.w.Par, ascending wing of parasphenoid; Bsph, basisphenoid; bucc.can, buccal canal; gr.j.v, groove for jugular vein; L.e, lateral ethmoid; L.r, lateral rostral; Na, nasal; Pa.a, anterior parietal; Pa.p, posterior parietal; Par, parasphenoid; pi, pillar; pr.con, processus connectens; sph.c, sphenoid condyle; So, supraorbital series; so.s.c, supraorbital sensory line canal; v.l.fo, ventrolateral fossa; ?v.pr.Pa, ventral (descending) process of the parietal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g002

The postparietal shield of the skull is divided medially into two of the supratemporal sensory line canal commissure. However, the strongly flattened halves (Figures 5, 6, 7, 8). Although it is slightly same interpretation is difficult to make because of the poor broken posteriorly and medially, the postparietal shield is shorter preservation of the surface of the skull roof. The anterior portion than the parietonasal shield. The anterior tip of the postparietal is of the postparietal presents a semi-circular depression (Pp.a.d, thick and narrow, leaving no gap between the two halves when Figure 7) as in Swenzia [8]. The suture between the postparietals they are put in contact. In its anteriormost portion, the and supratemporals runs anteroventrally on both sides (Figures 5, postparietal shield is thus as broad as the parietonasal shield. It 7). It extends far ventrolaterally through the third opening of the is most probable that the postparietal shield was strongly vaulted sensory line canal and terminates at the level of its ventral margin: and broad over most of the otoccipital portion as in Macropoma, such a condition is also observable in Libys (figure 3.17 in [12]). Libys, Holophagus [12] and Swenzia [8], and narrow anteriorly. The The supratemporal (Stt, Figures 5, 6, 7) is relatively large and center of ossification of the postparietal is situated very close to the represents almost half of the surface of the postparietal shield. On joint margin (Figures 6, 8), as in Libys [12]. The anterior thickening the left side of the skull, the lateral extrascapular (Ext.l, Figures 5, of the postparietal bears a concave facet (Figure 8), which probably 6, 7, 8) is present, dorsal to the suture between the postparietal and matched the contour of the parietal descending process. The the supratemporal. ventral surface of the anterior thickening of the postparietal 1.2 Sensory line canals. The otic sensory line canal opens exhibits a ridge that is oriented anteroposteriorly and flanked by through remarkably large vacuities along the supraorbital series tiny pores. The dorsal surface of the postparietal (Pp, Figures 5, 7) and flanks the parietonasal series laterally (so.s.c, Figures 2, 3). The shows well-marked longitudinal grooves, especially in its posterior ethmosphenoid portion of the skull bears nine vacuities and the portion. Similar grooves are also present in Macropoma, Latimeria otoccipital portion five. Slender pillars (pi, Figures 2, 3, 5, 7) [12] and Swenzia [8], and were interpreted as the anterior branches separate adjacent vacuities. Several pillars display a clear

PLOS ONE | www.plosone.org 4 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 3. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Ethmosphenoid portion of the skull in left lateral view. Abbreviations: ant.pr, antotic process; a.w.Par, ascending wing of parasphenoid; Bsph, basisphenoid; bucc.can, buccal canal; f.v.nas-b.can, foramen for ventral branch of naso-basal canal; gr.j.v, groove for jugular vein; io.s.c, infraorbital sensory line canal; L.e, lateral ethmoid; L.r, lateral rostral; Na, nasal; nos.a, anterior nostril; nos.p, posterior nostril; Pa.p, posterior parietal; Par, parasphenoid; pi, pillar; pr.con, processus connectens; sph.c, sphenoid condyle; So, supraorbital series; so.s.c, supraorbital sensory line canal; v.l.fo, ventrolateral fossa; v.pr.L.r, ventral (descending) process of the lateral rostral; ?v.pr.Pa, ventral (descending) process of the parietal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g003 separation between their dorsal edge and the parietonasal shield. in the orientation of the pillars is observable along the antero- This could be attributed to the lateral compression, or it could posterior axis (Figures 2, 3). The first three pillars are oriented suggest that most of the pillars may have been sutured to the anterodorsally whereas the more posterior ones are oriented parietonasal shield. We support the latter assumption based on the posterodorsally. The supratemporal bears one pillar, the posterior presence of clear suture between the anteriormost pillar and the parietal four, the anterior parietal two, and the anteriormost ones lateral rostral on both side of the skull. Consequently, we identify are borne by the nasals. A vacuity occurs between the edges of the pillars as expansions of the supraorbital series. The condition adjacent bones of the parietonasal series (i.e. posterior parietal/ observed in Megalocoelacanthus is thus very similar to that of the anterior parietal; anterior parietal/nasal). The suture between the Jurassic genus Libys, where the pillars forming the large vacuity of bones of the parietonasal series extends through the vacuity and is the supraorbital sensory line canal are interpreted as elements of subsequently overlapped by the ventral edge of the cavity the supraorbital series (Dutel pers. obs. on BMNH P.3337). A twist (Figure 2). This suggests that the parietonasal series could have

PLOS ONE | www.plosone.org 5 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 4. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Isolated snout. A, right anterolateral view; B, posterior view. Abbreviations: ant.ros, anterior opening for the rostral organ; c.nos.a, canal for the anterior nostril; d.l.Pmx, dorsal lamina of the premaxilla; f.a.n.c, anterior foramen of the nasal capsule; L.r, lateral rostral; m.p.s.c, median pore for the sensory line canal; n.c, nasal capsule; nos.a, anterior nostril; Pmx, premaxilla. Scale bar = 1 cm. doi:10.1371/journal.pone.0049911.g004

Figure 5. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Otoccipital portion of the skull in right lateral view. Abbreviations: Ext.l, lateral extrascapular; ?f.o.a, foramen for the orbitonasal artery; ot.s.c, otic sensory line canal; ot.s.c.m, medial branch of the otic sensory line canal; ot.sh, otic shelf; pi, pilar; Pp, postparietal; Pro, prootic; Stt, supratemporal; Stt.com, supratemporal commissure; v.pr.Stt, ventral (descending) process of the supratemporal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g005

PLOS ONE | www.plosone.org 6 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 6. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Otoccipital portion of the skull in right medial view. Abbreviations: Ext.l, lateral extrascapular; ot.sh, otic shelf; Pp, postparietal; Pro, prootic; Stt, supratemporal; v.pr.Pp, ventral (descending) process of the postparietal; v.pr.Stt, ventral (descending) process of the supratemporal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g006 extended laterally on both sides of the skull, so that it was interpreted as the median pore of the sensory line canal (m.p.s.c, overlapped by the supraorbital series. The first three vacuities are Figure 4A). Forey [12] suggested that this opening was related to more dorsal in position than the posterior one. Their shapes are the ethmoid commissure. Following the interpretation of Forey also different along the antero-posterior axis (Figure 2): the [12], the presence of such a pore in Megalocoelacanthus could suggest anteriormost vacuities are dorsoventrally flattened while the the presence of an ethmoid commissure running beneath the posteriormost are anteroposteriorly flattened. This suggests that bones, as in Latimeria [19] and Macropoma [12], and emitting canals on either side, the branches of the sensory line canal were oriented that are directed towards the tip of the snout as in Latimeria. anteromedially along the ethmosphenoid portion of the skull, In the otoccipital portion, the sensory line canal passes within meeting dorsomedially behind the lateral rostral to form the the postparietal and the supratemporal (Figures 5, 7). The antorbital sensory line canal commissure (ant.com.so.s.c, Figure 2). vacuities present on the otoccipital portion of the skull are larger The position of the antorbital commissure is thus at the same level than those of the ethmosphenoid portion and nearly square. The along the shield as in Latimeria [19] and Macropoma (figure 3.18 in posteriormost vacuity of the otoccipital portion opens throughout [12]). the supratemporal. It is triangular and that of the left side is The course of the sensory line canal along the anteriormost pierced by two foramina that are aligned along the anteroposterior portion of the skull is more difficult to reconstruct. The lateral axis (Figure 7). A slight swelling on the dorsal surface suggests that rostral (L.r, Figures 2, 3) shows no pores on its dorsal surface, as in the otic sensory line canal, lateral sensory line canal and Diplurus, Macropoma, Latimeria, Laugia [12], Swenzia [8] and supratemporal commissure (Stt.com, Figures 5, 7) met near the Rhabdoderma [17]. Consequently, the course of the infraorbital posterior edge of the supratemporal, as in Libys. A few pits lie on sensory line canal cannot be followed throughout the surface of the the dorsal surface of the postparietal (ot.s.c.m, Figures 5, 7), close bone. Megalocoelacanthus presents, as in Macropoma, Rhabdoderma, and to the joint margin. As in Holophagus (figure 3.18 in [12]), Libys Whiteia, a small median opening between the premaxillae that was (figure 3.17 in [12]), Macropoma (figure 3.21 in [12]) these pits could

PLOS ONE | www.plosone.org 7 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 7. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Otoccipital portion of the skull in left lateral view. Abbreviations: Ext.l, lateral extrascapular; ot.s.c, otic sensory line canal; ot.s.c.m, medial branch of the otic sensory line canal; ot.sh, otic shelf; pi, pillar; Pp, postparietal; Pp.a.d, anterior depression of the postparietal; Pro, prootic; Stt, supratemporal; Stt.com, supratemporal commissure; v.pr.Stt, ventral (descending) process of the supratemporal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g007 be related to the medial branch of the otic sensory line canal. Pit Macropoma [12]. The shallow ventrolateral fossa (v.l.fo, Figures 2, lines are not observed on the postparietal. 3) is clearly marked by an oblique, dorsal depression. 1.3 Neurocranium. The neurocranium of Megalocoelacanthus The basiphenoid is preserved in connection with the whole is extremely well preserved although strongly flattened laterally. It ethmosphenoid portion of the skull in AMNH FF 20267 (Figures 2, is extensively ossified, and completely divided into ethmosphenoid 3), and as an isolated element in the holotype specimen CCK 88- (Figures 2, 3, 4) and otoccipital portions (Figures 5, 6, 7, 8), which 2-1 (Figure 9). In AMNH FF 20267, the basisphenoid is relatively are articulated through a transversally straight intracranial joint. large compared to the entire ethmosphenoid portion of the skull. The basisphenoid of the holotype specimen CCK 88-2-1 (Figure 9) The paired processus connectens (pr.con, Figures 2, 3, 9) are is isolated and well preserved. Anterior and posterior catazygals robust, well developed, and posteriorly elongated. The angle are preserved in AMNH FF 20267 (Figure 10). between the parasphenoid and the processus connectens is smaller The ethmosphenoid portion (Figures 2, 3) is robust, longer than than what can be observed in Latimeria and Macropoma (Figures 2, the otoccipital portion, but its anterior end is broken at the level of 3). The surface of the processus connectens is rugose and was the anterior margin of the lateral rostral. It consists of a large probably capped by cartilage to articulate with the otic shelf of the basisphenoid tightly sutured to the parasphenoid, and a pair of prootic. The area between the processus connectens and the lateral ethmoids. The anterior face of the lateral ethmoid (L.e, antotic process is marked by a deep groove (gr.j.v, Figures 2, 3, 9A) Figures 2, 3) is roughened and sutured to the ventral process of the that houses the jugular vein in Latimeria [18]. The antotic process lateral rostral. Posterior to this area the lateral ethmoid is notched (ant.pr, Figures 2, 3, 9A, 9B, 9C) is prominent, oriented to form the ventral margin of the buccal canal (bucc.can, Figures 2, anteroventrally, and covered with strong ridges in AMNH FF 3). The lateral ethmoid enlarges dorsally towards its posterior end, 20267 on which may have been anchored the adductor palatini which is partially overlapped by the lateral rostral. A large muscle. In AMNH FF 20267 and CCK 88-2-1 the anterodorsal foramen (f.v.nas-b.can, Figure 3), interpreted as the point of part of the antotic process is notched and marks the posterior emergence of the ventral branch of the naso-basal canal, pierces margin of the suprapterygoid fossa (spt.fos, Figures 2, 3, 9B, 9C). the lateral ethmoid beneath the contact with the ventral process of No trace of foramina can be observed on both specimens, the lateral rostral. Such a foramen is also described in Undina and probably due to the artefact of the strong lateral compression of

PLOS ONE | www.plosone.org 8 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 8. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Otoccipital portion of the skull in left medial view. Abbreviations: Ext.l, lateral extrascapular; ot.sh, otic shelf; Pp, postparietal; Pro, prootic; v.pr.Pp, ventral (descending) process of the postparietal. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g008 the bones. Although both basisphenoids are flattened laterally, the [12]), rather than to that of Mawsonia or Axelrodichthys where the one of the holotype specimen CCCK 88-2-1 (Figure 9) is less so sphenoid condyles are very close medially and separated by a deep and enables further comparisons with other coelacanths. The notch (figures 1A, 18A in [5]). basisphenoid of Megalocoelacanthus is higher and narrower than that Two halves of the otoccipital portion are preserved, but of Axelrodichthys, Mawsonia and Diplurus [5,20]. The dorsum sella extremely flattened (Figures 5, 6, 7, 8). The prootic (Pro, (d.s, Figure 9) is much elongated in Megalocoelacanthus compared Figures 5, 6, 7, 8) is short, ventrally oriented, and shows a very with these genera. As in other coelacanths [12], the basisphenoid short and narrow otic shelf whose inner surface is covered by tiny was certainly pierced medially by the buccohypophyseal canal rugosities (ot.sh, Figures 6, 8). It is enlarged posteriorly and its (?pit.fos, Figure 9C). However, it fails to pierce the ventral surface posterior end is broken on both sides. The inclination of the otic of the parasphenoid in AMNH FF 20267 (Figure 11B), contrary to shelf is less prominent than in Macropoma, but this may be due to what is observed in basal coelacanths such as Diplocercides, deformations of the specimen. Euporosteus [12], Miguashaia [21], and Styloichthys [22,23]. This The prootic presents two roughened areas. The anterior one condition observed in Megalocoelacanthus seems to be derived among (the so-called prefacial eminence) is sutured on the inner side with coelacanths and also observed in Axerodichthys (AMNH 14026L), the postparietal descending process (v.pr.Pp, Figures 6, 8), and the Latimeria (MNHN C24), Macropoma (figure 6.10 in [12]), Undina posterior one is sutured with the supratemporal descending (BSPG 1870 XIV 508). In posterior and dorsal views, the process (v.pr.Stt, Figures 5, 6, 7). This latter suture can be basisphenoid of the holotype specimen CCK 88-2-1 presents an observed only on the lateral part of the right side (Figure 5). overlapping surface for the descending process of the parietal Between these two areas, temporal excavation is marked by a (o.v.pr.Pa, Figure 9B). The sphenoid condyles (sph.c, Figure 9A, slight concavity on both sides (Figures 5, 7), and seems to be lined 9C, 9D) are well separated by the concave posterior margin of the with bones like in Axelrodichthys and Mawsonia [4]. The condition basisphenoid. The condition observed in Megalocoelacanthus is very observed in Megalocoelacanthus is thus different from that of Latimeria similar to that of Latimeria [18] and Macropoma (figure 6.12B, D in and Macropoma, where the temporal excavation is cartilaginous. In

PLOS ONE | www.plosone.org 9 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 9. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, holotype specimen CCK 88-2-1 from lower Campanian of the Blufftown Formation. Isolated basisphenoid. A, right lateral view; B, posterior view; C, anterior view; D, dorsal view. Abbreviations: ant.pr, antotic process; d.s, dorsum sellae; gr.j.v, groove for jugular vein; n.p, notochordal pit; o.v.pr.Pa, overlapping surface for descending process of parietal; pit.fos?, pituitary fossa; pr.con, processus connectens; sph.c, sphenoid condyle; spt.fos?, suprapterygoid fossa. Scale bar = 5 cm. doi:10.1371/journal.pone.0049911.g009 the inner part of both sides (Figures 6, 8), the suture between the contact with the supratemporal descending process. The identi- descending process of the postparietal and the prootic runs fication of such foramina is very difficult in Megalocoelacanthus. Most anteroventrally from the dorsal edge of the otic shelf to the probably, evidence for a foramen could be found on the lateral anterior margin of the otoccipital portion. The path of this suture face of the right moiety of the prootic (?f.o.a, Figure 5). Like in is unclear on the lateral part of both sides because it is completely Macropoma and Mawsonia a foramen opens ventrally to the overlapped by the postparietal shield on the right side, and only roughened area contacting the supratemporal descending process, the anteriormost portion of the suture can be observed on the left. which allows us to suggest that it could correspond to the foramen However, it seems that the suture is directed posterodorsally on the for the orbital artery. lateral side. The contact between the prootic and the descending Posterior to the otic shelf, the lateral wall extending postero- process of the postparietal (Figure 5) appears to be much more ventrally is broken on both sides of the skull. The saccular ventral that what can be observed in other coelacanths, but this is chamber is completely flattened between the postparietal shield probably an artefact due to the strong deformation of this portion and the medial wall that usually separates it from the notochordal of the skull. canal. The course of the arteries and nerves through the prootic is In coelacanths, the base of the otoccipital portion of the quite variable among coelacanths [12]. In Macropoma, the lateral neurocranium is poorly ossified and composed of several elements surface of the prootic is pierced by several foramina that are that embed the notochord ventrally. The basioccipital alone is lacking in Latimeria: the palatine nerve emerges ventrally to the sutured to the posterior wing of the prootic, whereas the anazygal, prefacial eminence, and the orbital artery opens ventral to the and the anterior and posterior catazygals occupy the basicranial

PLOS ONE | www.plosone.org 10 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

rises steeply, so that its posteriormost dorsal surface reaches the anterior margin of the processus connectens. This condition is unknown in Macropoma, Latimeria (figures 6.1, 6.11 in [12]) and other Mesozoic coelacanths such as Axelrodichthys (Maisey 1986), but is present in Rhabdoderma (figure 6.5 in [12]). The parasphenoid ends posteriorly abruptly below the anterior side of the processus connectens, with a slightly curved ventral margin. The palatoquadrate (Figure 11A) consists of the pterygoid anteriorly, the metapterygoid posterodorsally, and the quadrate posteroventrally. The palatoquadrate is triangular in shape, short and very deep. Its general shape is thus proportionally similar to that of Latimeria, Macropoma, and Holophagus [12], whereas Mawsonia and Axelrodichthys possess a longer and shallower palatoquadrate [5]. The anteriormost part of both pterygoids (Pt, Figure 11A) of AMNH FF 20267 as well as that of the holotype specimen CCK 88-2-1 (figure 2F in [1]) is thin and covered with striations. In these specimens as well as in AUMP 3834, and FMNH P27524 (Schwimmer pers. obs.) the anterior termination is thus similar, suggesting that it was poorly ossified or capped by a cartilage layer that was connecting it to the autopalatines which are preserved separately in AMNH FF 20267 (Figure 11C). The pterygoid (Pt, Figure 11A) is triangular, shallow and short, and presents a ventral swelling (v.sw.Pt, Figure 11A) anterior to the quadrate that is more pronounced than that of Macropoma or Latimeria. The pterygoid forms a narrow and straight edge along the anterior margin of the metapterygoid. On the medial side of the palatoquadrate, the pterygoid also overlaps the metapterygoid and forms its posterior edge. The medial surface of the pterygoid is ornamented with tubercular shagreen. The metapterygoid (Mpt, Figure 11A) is short but very large compared to that in other coelacanths. As in Latimeria, it is saddle- shaped and its dorsal surface was articulating with the antotic process (art.ant.pr, Figure 11A). When considering the entire Figure 10. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the palatoquadrate, its size is proportionally quite small. Contrary to Niobrara Formation. A, anterior catazygal. B, posterior catazygal. what can be observed in Mawsonia or Axelrodichthys, the metapter- doi:10.1371/journal.pone.0049911.g010 ygoid displays a marked ventral recess ventral to the dorsal edge of the pterygoid on the medial side. fenestra and lie free from the rest of the neurocranium. In The quadrate (Q, Figure 11A) is straight vertically and extends Latimeria, these elements are firmly attached to the neurocranium dorsally up to the level of the ventral half of the pterygoid. It by mean of strong ligaments (Dutel pers. obs. on MNHN C24). finishes dorsally as an open-end prolonged by an anterodorsally The anterior and the posterior catazygals are preserved in AMNH curved ridge extending onto the pterygoid and metapterygoid. FF 20267 (Figure 10). The anterior catazygal (Figure 10A) is the This suggests the presence of a posterior cartilage joining the largest one, semi-circular in shape, wider than long, with concave dorsal end of the quadrate to the metapterygoid. The double lateral margins. The posterior catazygal (Figure 10B) is much condyle of the quadrate is large and robust, as in Latimeria and smaller, longer than wide, and bell-shaped in dorsal/ventral views. Mawsonia. Its anterior margin is straight whereas the posterior is rounded and 1.5 Lower jaw, coronoids and gular plates. Both lower narrower. Although these elements are rarely preserved in fossil jaws are well preserved (Figures 12, 13). The lower jaw is long and coelacanths, they are well known in Mawsonia [2,3]. Anterior shallow throughout, and resembles that of Undina, Holophagus, catazygals referred to this genus are butterfly-shaped whereas the Latimeria and Macropoma. The dentary (De, Figures 12A, 13A) is posterior catazygal is semi-lunar in shape [2]. Catazygals are long and narrow, and its proportion relative to the total jaw length unossified in Axelrodichthys [2]. The catazygals described in is close to that seen in Undina and Holophagus, where it reaches Megalocoelacanthus are very different to that of Mawsonia, but rather about 40% of the total jaw length [12]. The dentary overlaps the resemble those of Holophagus (figure 6.9 in [12]) Macropoma angular and possesses a hook-shaped process extending poster- (figure 6.10 in [12]) and Latimeria [18]. odorsally. The distal part of this hook-shaped process is broken, suggesting that it may have been more prominent and elongated, 1.4 Palate. The parasphenoid (Par, Figures 2, 3, 11) is very comparable to that of Libys, Undina, Holophagus and Macropoma. The narrow and deep. As in Latimeria, its anterior half is marked by low splenial is narrow, and slender. It forms the ventral edge and and elongated lateral wings (a.w.Par, Figures 2, 3), which are ventrolateral side of the anterior portion of the jaw, and extends connected to the ventral margin of the lateral ethmoid. In ventral posteriorly to the hook-shaped process of the dentary (Figures 13A). view (Figure 11), the parasphenoid is strongly compressed laterally It presents no traces of ornamentation on its surface. Ventrally, the along its posterior half, and expands on both sides at the level of splenial (Spl, Figures 12, 13) forms a hump that winds around the the anterior half. The anterior portion is ovoid in shape, ventrally ventral edge of the mentomeckelian (Mm, Figures 12B, 13B). The concave and covered by tiny villiform teeth (t.Par, Figure 11). In mentomeckelian forms the anteriormost part of medial side of the lateral views (Figures 2, 3), the posterior half of the parasphenoid jaw (Figures 12B, 13B). It is rectangular in shape, with a posterior

PLOS ONE | www.plosone.org 11 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 11. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Palate bones. A, right palatoquadrate in lateral (top) and medial (bottom) views. B, parasphenoid in ventral view. C, right (top) and left (bottom) autopalatines in lateral (left) and medial (right) views. Abbreviations: art.ant.pr, surface for articulation with the antotic process; Bsph, basisphenoid; Mpt, metapterygoid; n.p, notochordal pit; Par, parasphenoid; pr.con, processus connectens; Pt, pterygoid; Q, quadrate; t.Par, toothed area of parasphenoid; v.sw.Pt, ventral swelling of pterygoid. Scale bar = 5 cm. doi:10.1371/journal.pone.0049911.g011 finger-like expansion ventrally to the prearticular (Part, that forms the coronoid process (co.pr.Ang, Figures 12, 13). The Figure 12B). It is inwardly curved, so that its anteriormost portion dorsal margin of the angular is concave posteriorly, and then runs overlaps the splenial in lateral view (Figures 12A, 13A), to form the straight anteriorly up to the blunt process. This condition is very mandibular symphysis. different from that of the latimeriids Latimeria, Macropoma, and The angular (Ang, Figures 12A, 13A) is triangular in shape and Swenzia where the dorsal edge of the angular is convex and regular slightly concave ventrally. It is deeper than in Holophagus, Latimeria, throughout, and of the mawsoniids Mawsonia and Axelrodichthys Macropoma, Undina, and Swenzia. Just behind its contact with the where it forms a deep hump. The center of ossification of the dentary, it shows a prominent dorsal, anteriorly bent extension angular can be observed at the level of its deepest portion. A long

PLOS ONE | www.plosone.org 12 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 12. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Right lower jaw. A, lateral view. B, medial view. Abbreviations: Ang, angular; art.Q, surface for articulation with the quadrate; Co, coronoids; co.pr.Ang, coronoid process of the angular; De, dentary; d.p, enlarged sensory pore within the dentary; f.pop.s.c, opening for the preopercular sensory line canal; f.V.m, foramen for the mandibular ramus of the trigeminal nerve; f.VII.m.ext, foramen for the external mandibular ramus of the facial nerve; f.VII.m.int, foramen for the internal mandibular ramus of the facial nerve; i.add.md, insertion point for the adductor mandibulae muscle; i.art-hy.lig, insertion point for the articular-hyomandibular ligament; i.surf-a.intermd, insertion surface for the anterior ramus of intermandibular muscle; Mm, mentomeckelian; m.s.c, mandibular sensory line canal; Part, prearticular; Rart, retroarticular; ?sop.br, subopercular branch of the preopercular canal; Spl, splenial. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g012 and broad surface for the insertion of the anterior and posterior process of the dentary is elongated and closely associated with the ramus of the intermandibular muscle is seen on the ventral margin dentary. The surface of the coronoid is covered with shagreen of lower jaw (i.surf-a.intermd, Figures 12A, 13A) as in Latimeria.No tubercles, and does not bear any enlarged teeth. Anteriorly to the oral pit lines are observed on the angular. hook-shaped process of the dentary, only a smaller coronoid is The coronoid series is poorly preserved and few elements of the preserved. However, the rugous dorsal surface of the dentary series are only observed on the medial side of the right lower jaw suggests that the coronoid series were extending anteriorly up to (Co, Figure 12B). The coronoid posterior to the hook-shaped the level of the mentomeckelian. The left principal coronoid

PLOS ONE | www.plosone.org 13 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 13. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Left lower jaw. A, lateral view. B, medial view. Abbreviations: Ang, angular; art.Q, surface for articulation with the quadrate; co.pr.Ang, coronoid process of the angular; De, dentary; d.p, enlarged sensory pore within the dentary; f.pop.s.c, opening for the preopercular sensory line canal; f.V.m, foramen for the mandibular ramus of the trigeminal nerve; f.VII.m.ext, foramen for the external mandibular ramus of the facial nerve; gr.VII.m.int, groove for the internal mandibular ramus of the facial nerve; i.add.md, insertion point for the adductor mandibulae muscle; i.art-hy.lig, insertion point for the articular-hyomandibular ligament; i.surf-a.intermd, insertion surface for the anterior ramus of intermandibular muscle; Mm, mentomeckelian; m.s.c, mandibular sensory line canal; Rart, retroarticular; ?sop.br, subopercular branch of the preopercular canal; Spl, splenial. Scale bar = 10 cm. doi:10.1371/journal.pone.0049911.g013

(Figure 14A) is preserved as an isolated element. Its general shape medial surface of this coronoid is covered by tiny, villiform teeth is rounded with a broad base that extends posteriorly as a finger- that form a shagreen area. like process. The dorsal portion of the principal coronoid is curved The prearticular is preserved only on the right jaw (Part, and much longer than in Macropoma and Libys where it is developed Figure 12B). It consists in a slender, long and shallow bone that as a narrow dorsal process [12]. Deep furrows mark the entire covers most of the medial side of the lower jaw, and like the length of its ventrolateral edge. Digitations are present on the principal coronoid, is covered with tiny villiform teeth. The posterodorsal edge, from which long grooves extend radially. The articulars (Figure 14B) are preserved as isolated elements and consist of a single ossification. They possess two concave

PLOS ONE | www.plosone.org 14 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 14. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. A, left principal coronoid in lateral (left) and medial (right) views. B, articulars of the left (top) and right (bottom) lower jaws. Scale bar = 2 cm. C, gular plates in external (left) and internal (right) views. Abbreviation: i.rd.intermed: insertion ridge for the intermandibular muscle. Scale bar = 10 cm. Arrows oriented anteriorly. doi:10.1371/journal.pone.0049911.g014 articulatory facets, one dorsal and one ventral that are anterome- as well as on holotype CCK88-2-1, suggesting that the main dially inclined. The retroarticular (Rart, Figures 12B, 13B) mandibular sensory line canal runs down from the preopercular possesses a double articulatory facet that faces the articular. and enters the angular at this level, as in Latimeria [18,19,24,25] Posteriorly, the retroarticular bears a small facet for the and Macropoma [12]. Anteriorly, a long and slender foramen lies articulation of the symplectic. As in other coelacanths, the along the suture between the splenial and the dentary at the level posterior portion of the retroarticular may have been capped of the hook-shaped process of the dentary (dp, Figures 12A, 13A). with cartilage. A ridge interpreted as the insertion point for the Such a foramen is also present in the same position in Macropoma, articular-hyomandibular ligament is observable on the lateral Holophagus and Undina [12]. This foramen could correspond in surface of the retroarticular (i.art-hy.lig, Figure 12). Longitudinal Latimeria to the enlarged pore of the sensory line canal that pierces ridges are present on the medial surface of the retroarticular. the splenial just beneath the suture with the dentary. This pore is The course of the main sensory line canal in Megalocoelacanthus connected to the main mandibular canal that runs through the seems to be very similar to that observed in Latimeria and angular. In Macropoma and Holophagus, the subopercular branch of Macropoma. The mandibular sensory line canal is a large canal that the preopercular canal exits posteriorly on the angular immedi- opens ventrally through large pores on the surface of the angular ately beneath the foramen for the external ramus of the and splenial (m.s.c, Figures 12A, 13A). The pores are oriented mandibular branch of the facial nerve [12]. Such a foramen has ventrally on both bones. Four pores are clearly observed on the not been observed in this area in Megalocoelacanthus. If the splenial and five on the angular. The anteriormost pores open on subopercular branch of the preopercular is present in Megalocoe- the angular are oriented anteroventrally whereas the posterior lacanthus, the posterior orientation of the posteriormost pore in the ones are oriented posteroventrally. Posteriorly, a marked notch is angular would mark its point of exit (?sop.br, Figures 12A, 13A). present at the margin of the angular, lateral and posterior to the This condition would thus be similar to that in Latimeria [18]. articular glenoid. A foramen can be clearly observed in this area Anterior to the enlarged pore of the sensory line canal, another on the right mandible of AMNH FF 20267 (f.pop.s.c, Figure 13B) foramen pierces the dentary just above the suture with the splenial

PLOS ONE | www.plosone.org 15 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

(f.V.m, Figures 12A, 13A). As in Latimeria, this foramen may directed elbow-like expansion is hook-shaped like in Axelrodichthys correspond to the exit of the mandibular branch of the trigeminal (AMNH 13962 R), Latimeria [18], Macropoma [12] and Mawsonia nerve for innervating the skin of the lower lip. The relative position (AMNH 11758). The ventral expansion is slender and slightly of the enlarged pore of the sensory line canal and the foramen for convex, and its posterior portion bears a groove that may have the mandibular branch of the trigeminal nerve differs from that of been capped with cartilage, as well as the anterior margin of the Latimeria. In this genus the enlarged pore of the sensory line canal bone that finishes in a dead-end. In AMNH FF 20267 as well as in is ventral to the foramen for the mandibular branch of the the holotype specimen CCK 88-2-1 (figure 2L in [1]), the trigeminal nerve, whereas both are aligned antero-posteriorly ceratohyal is much more curved, and its ventral expansion appears along the dentary-splenial suture in Megalocoelacanthus. to be shorter than that of Axelrodichthys (FMNH FM11856), A foramen for the external mandibular ramus of the facial nerve Latimeria [18], and Mawsonia (AMNH 11758). The dorsal end and (f.VII.m.ext, Figures 12A, 13A) is present on the lateral surface of the elbow-like expansion are broken, but they are usually also the jaw between the retroarticular and the angular. In Latimeria, capped by cartilage. the internal ramus of the mandibular branch of the facial nerve 1.8 Branchial arches, urohyal. The branchial skeleton is penetrates the mandible by a foramen situated between the poorly known in fossil coelacanths. The branchial skeleton of prearticular and the retroarticular. In Megalocoelacanthus a ventral Megalocoelacanthus consists of a complete urohyal, and a basibran- groove can be observed on the anterior margin of the retro- chial associated with tooth plates and ceratobranchials. articular, at the level of the articulation facet (Figures 12B, 13B). The basibranchial (Bb, Figure 16A) is large and rounded in The absence of the prearticular on the left lower jaw shows the shape. It is fused with three tooth plates (p.t.p.Bb, a.t.p.Bb, course of this groove within the mandible (gr.VII.m.int, Figure 16A). As in Macropoma [12], the basibranchial consists in a Figure 13A). Forey [12] also observed such a groove in Macropoma central embedded ossification surrounded by a perimeter of and interpreted it as the mark of the path of the internal cartilage. Its ventral surface is marked by a posterior median pit mandibular ramus of the trigeminal nerve within the mandible. for the articulation of the urohyal (art.Uhy, Figure 16A), and by Both gular plates are preserved with little deformation paired lateral pits that were probably articulated with the first two (Figures. 14C). Their shape is very similar to those of Latimeria, ceratobranchials as in Latimeria and Macropoma (art.Cb1, art.Cb2, and their surface is slightly concave dorsally. Their anterolateral Figure 16A). Tooth plates associated with the basibranchial consist edge is slightly swollen and the anterior tips diverge strongly along in an anterior median diamond-shape plate (a.t.p.Bb, Figure 16A) the posterior midline. The lateral edge of the gular plates is curved and a pair of large plates, extending symmetrically posteriorly in its posterior half, so that the posterior tips meet medially. On the along the midline (p.t.p.Bb, Figure 16A). The ventral surface of the internal surface of the gular plate, a ridge runs along the latter displays a paired concavity that is interpreted for the anteroposterior axis parallel to the lateral edge (i.rd.intermd, articulation of the ceratohyal (art.Ch, Figure 16A). Tiny villiform Figures. 14C). In Latimeria, this ridge corresponds to the insertion teeth cover all the tooth plates. Although basibranchial tooth point of the anterior and posterior ramus of intermandibular plates are rarely found in situ, previous studies support that the muscle (Dutel pers. obs. on MNHN C24). Neither their internal trend in basibranchial tooth plate evolution in coelacanths was nor their dorsal surfaces are ornamented and the gular pit-lines are towards a mid-line fusion of many paired plates into larger tooth not observed. plates [12,26]. The pattern observed in Megalocoelacanthus is also 1.6 Cheek bones and opercular. Cheek bones are poorly observed in Latimeria and Undina where the tooth plates associated preserved in AMNH FF 20267. An isolated element presenting a with the basibranchial consist of large paired plates posteriorly to a sensory line canal crossed by pillars could be interpreted as a smaller median plate, while Diplurus and Axelrodichthys bear, fragment from the lachrymojugal or the postorbital. This suggests respectively, three and two pairs of large plates [12]. that the sensory line canal was opening through cheek bones by The urohyal (Figure 16B) is very well preserved and shows no large vacuities as on the skull roof. Consequently, the condition in evidence of strong deformation. The general shape of the urohyal Megalocoelacanthus would have been identical to that of Libys were is characteristic of what can be observed in other coelacanths. The the large sensory line canal is opening through a large continuous bifid posterior end is broad and form two semi-lunar wings. They groove crossed by pillars on the lachrymojugal and postorbital arise at the level of the first third of the bone, making them more (BSPG 1860 XIV 502). Like in this genus and other latimerioids prominent than those of Latimeria and Macropoma. The slit between [12], it is probable that cheek bones were well separated from each the posterior wings is straight, narrow and prolonged anteriorly by other in Megalocoelacanthus, explaining the poor preservation of this a broad groove. These conditions are thus different from that complex. observed in Axelrodichthys (FMNH FM11856) in which the lateral Both operculars are preserved (Figure 15D). They are partly wing extends straightly and the slit is V-shaped. The slit is more broken posteriorly and ventrally, but seem to be deeper than long expended anteriorly than that of Macropoma (figure 7.7 in [12]) and as in Latimeria, Holophagus, and Macropoma. Their center of Latimeria (figure 7.6 in [12]). The ventral surface of the urohyal is ossification is situated anterodorsally, and both their lateral and flat, except on its anterior portion where the edges are slightly medial surfaces are ornamented by isolated tubercles. The raised. The anterior end is also bifid with a hump on each side. anterodorsal edge of the operculars is notched and similar in The dorsal surface of the posterior wings is concave in its anterior shape to that of Macropoma and Holophagus [12]. portion and flat in its posterior portion. A long and deep septum 1.7 Hyoid arch. The hyoid arch of coelacanths consists of covered by small grooves extends medially. Although some notable hyomandibular, interhyal, ceratohyal, hypohyal and symplectic. differences are observed, the general shape of the urohyal of Only one ceratohyal and the left symplectic are preserved here. Megalocoelacanthus is much more similar to that of the latimeriids The left symplectic (Figure 15B) is entirely preserved but Macropoma and Latimeria, than to that of the mawsoniids Mawsonia strongly compressed laterally. Its shape is typical, with an upper and Axelrodichthys. part enlarged posteriorly. As in other coelacanths, both ends were As usual in fossil coelacanths, little remains of the gill arches. probably cartilaginous and its actual length may have been longer. Four ceratobranchials are preserved (Figure 15A), and two of them The ceratohyal (Figure 15C) has a typical shape for coelacanth. are almost complete. However, it is difficult to determine their It is laterally compressed, curved posteriorly, and its posteriorly position in the branchial series. Ceratobranchials are curved and

PLOS ONE | www.plosone.org 16 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 15. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. A, branchial arches; B, symplectic; C, ceratohyal; D, left (top) and right (bottom) operculars in lateral (left) and medial (right) views. Scale bar = 5 cm. doi:10.1371/journal.pone.0049911.g015 compressed laterally. The ventral extremity is rounded and was The most remarkable feature is the very small size of the clavicle probably articulated with the basibranchial. The bone narrows (Figure 18C) compared to that of the cleithrum. As is typical, the dorsally into a sharp dorsal end. Isolated small, sharp, denticles clavicle twists medially from the leading edge of the cleithrum. The measuring about 1–2 mm are preserved in the matrix at the edge distal tip of the clavicle presents marked, irregular digitations, and of the ceratobranchial of the holotype specimen CCK 8-22-1 the contact with its antimere was probably made through cartilage (Figure 17). These denticles were recognized as branchial teeth by as in Latimeria [18]. Dorsally, the clavicle enlarges as a thin, Schwimmer [27] and we here follow this interpretation. When medially concave layer of bone that was overlapping the lateral compared to the size of the ceratobranchial, these denticles are side of the cleithrum. significantly smaller than those of Latimeria and Axelrodichthys Both scapulocoracoids are preserved, but strongly flattened (FMNH FM11856). (Figures 18D, E). Each consists of a single, ossified element with a 1.9 Postcranial skeleton and scales. Very few elements of broad and flat proximal portion articulated with the cleithrum, the postcranial skeleton are preserved. All axial skeleton and fins and a short distal portion bearing the glenoid surface for the first are missing, and only the pectoral girdle and some isolated scales axial mesomere. Although it is partially crushed, the best preserved are present. scapulocoracoid presents a concave glenoid surface, covered with The shoulder girdle (Figure 18) is narrow and is represented by coarse rugosities, and which may have been capped with cartilage the cleithrum, the extracleithrum, the clavicle and the scapulocor- in life (Figure 18E). acoid. The pectoral girdle of coelacanths usually includes one The axial skeleton of Megalocoelacanthus is only known from an supplementary dermal bone, the anocleithrum, which is not isolated and well preserved vertebra found on the holotype preserved here. specimen CCK 8-2-22 (Figure 19A). It is very similar to that of The cleithrum (Figures 18A, B) is compressed laterally and bent other coelacanths: the neural arch is forked and co-ossified with a anteriorly. The left cleithrum (Figure 18A) is the best preserved median neural spine. In coelacanths, the anterior neural spines are with the extracleithrum sutured on the lateral side, and only lacks short, and gradually increase in height posteriorly. Here, the its dorsal and ventral tips. It is elbow-shaped posteriorly and neural spine is bended posteriorly and relatively short compared to presents a ventral and dorsal half very distinct in shape. The latter those situated posteriorly to the D1 in other coelacanths. Thus, the is slender, and more developed and straight than in Latimeria [18], vertebra preserved here was most probably situated anteriorly whereas the former is broad and rounded posteriorly. Contrary to along the body axis. what can be observed in Mawsonia and Axelrodichthys [12], there is The best-preserved body scale of AMNH FF 20267 is no broad medial extension of the cleithrum, and its general shape subcircular, and is about 5 cm in diameter (Figure 19B). Only is more similar to what can be observed in Latimeria or Macropoma. its overlapped portion is preserved. As in other coelacanths, the exposed portion seems to represent here less than one third of the

PLOS ONE | www.plosone.org 17 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 16. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. A, basibranchial and basibranchial tooth plate in dorsal view (top), and ventral view (bottom). B, urohyal in dorsal view (top), and ventral view (bottom). Abbreviations: a.t.p.Bb, anterior tooth plate of the basibranchial; art.Cb1, surface for articulation with the first ceratobranchial; art.Cb2, surface for articulation with the second ceratobranchial; art.Ch, surface for articulation with the ceratohyal; art.Uhy, surface for articulation with the urohyal; Bb, basibranchial; p.t.p.Bb, posterior tooth plate of the basibranchial. Scale bar = 5 cm. doi:10.1371/journal.pone.0049911.g016 total surface of the scale. The scale shows concentric ridges and follow these scorings. Character 52 ‘‘sclerotic ossicles absent (0), presents no evidence of pore canal system or lateral line system. present (1)’’ was scored as ‘‘1’’ for Libys by Forey [12]. This character is here scored as ‘‘0’’ because the respective holotypes of 2. Phylogenetic analysis L. superbus and L. polypterus do not present sclerotic ossicles [29]. 2.1 Revised matrix with new characters and taxa. The Forey [12] and other authors [5,17,21,29] considered the visceral recent phylogenetic analyses investigating the relationships of calcified structure present in fossil coelacanths as a calcified swim coelacanths were made by Cle´ment [8], Friedman & Coates [13], bladder. However, recent histological studies on the structure of Yabumoto [14], Geng et al. [9], and Wendruff & Wilson [28] and the calcified organ in different specimens of Axelrodichthys and are based on the data matrix of Forey [12] with several corrections comparison with the fatty organ of the extant coelacanth Latimeria, and additions. Cle´ment (2005) corrected scoring for character 31 suggest that this is instead an ossified bladder with a respiratory ‘‘preopercular absent (0), present (1)’’ in the original Forey’s function rather than a buoyancy function [30]. Considering this matrix. Yabumoto (2008) subsequently changed Cle´ment’s scoring interpretation we modified character 107 ‘‘swim bladder not for Mawsonia to ‘‘1’’. According to Friedman & Coates [13], the ossified (0), swim bladder ossified (1)’’ into ‘‘ossified bladder absent state of character 54 ‘‘dentary teeth fused to the dentary (0)’’ (0), present (1)’’. We also reviewed the coding of this character for cannot be assessed for Allenypterus because its mandible is several taxa: it was coded ‘‘0’’ for Polyosteorhynchus, Allenypterus, and edentulous, and has to be scored as question mark. We here ‘‘?’’ for Mawsonia [12]. We here code this character ‘‘1’’ for these

PLOS ONE | www.plosone.org 18 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 17. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, holotype specimen CCK 88-2-1 from lower Campanian of the Blufftown Formation. Close-up view of branchial denticles present on the edge of the gill arches of the holotype specimen. Arrows indicate the denticles embedded in the matrix. Scale bar = 1 mm. doi:10.1371/journal.pone.0049911.g017 three genera based on Lund & Lund [17] and Brito et al. [30]. heuristic search was performed using the tree-bisection-reconnec- Character 23 ‘‘supraorbital sensory line canal opening through tion branch swapping algorithm (TBR) with 10,000 random bones as a single large pore (0), bifurcating pores (1), many tiny addition-sequence replicates. We ran the analysis with all pores (2)’’ was scored ‘‘0’’ for Libys, Latimeria, Diplurus, Laugia, and characters unweighted and multistate characters unordered. Whiteia. The supraorbital sensory line canal of Libys and Branches with a maximum length of zero were collapsed, so that Megalocoelacanthus is a continuous groove crossed by pillars that any branch supported by ambiguous synapomorphies is con- are probably formed by the supraorbitals. This condition is thus served. Bootstraps values were calculated with this program using very different from that of Latimeria (figure 3.1 in [12]), Diplurus heuristic searches and 1,000 bootstrap replicates, with 100 random (figure 4 in [31]), Laugia (figure 3.8 in [12]), and Whiteia (figure 3.15 sequence additions per replicate. Bremer decay indices were in [12]) and we do not consider it can be coded under the same calculated by combining PAUP 4.0b10 [34] and TreeRot v.3 [35]. state of character. Consequently, we integrated an additional state Optimizations of ambiguous states of characters were performed to character 23: ‘‘supraorbital sensory line canal opening through using the software WINCLADA 1.00.08 [36]. The tree is rooted bones as a single large pore (0), bifurcating pores (1), many tiny by two outgroups, porolepiforms and actinopterygians. pores (2), a large and continuous groove crossed by pillars (3)’’. Two analyses were carried out. The first analysis (Information Character 50 ‘‘Infraorbital, jugal and preopercular sensory canals S3) was run with all the taxa of the data matrix. The strict opening through many tiny pores (0), opening through a few large consensus tree (Figure 21) of the 584 equally parsimonious trees pores (1)’’ was coded ‘‘1’’ for Libys. However, in this genus (BSPG (length = 288; consistency index = 0.4132; retention index 1870 XIV 502) the infraorbital and preopercular sensory line = 0.6938) showed two areas of conflict within clade 2 and clade 3. canal displays the same condition as the supraorbital sensory The irresolution of the phylogenetic relationships is due to the canal, i.e. a large and continuous groove crossed by pillars. Based instability of Indocoelacanthus (clade 2), and Lualabaea (clade 3). The on an isolated element of the lachrymojugal, the same condition instability of these taxa has already been noted by Forey [12] and was probably present in Megalocoelacanthus. We thus propose an attributed to the high percentage of missing values (Indocoelacanthus, additional state for character 50 ‘‘Infraorbital, jugal and pre- 80%; Lualabaea, 99%) due to incompleteness of the fossil material. opercular sensory canals opening through many tiny pores (0), A second analysis was then conducted without these two taxa opening through a few large pores (1), a large, continuous groove (Information S4). The strict consensus tree is shown Figure 22, and stretched by pillars (2)’’. it will be used for presenting the following results. The diagnostic Additional characters and taxa are added in the matrix taken information for the nodes and terminal taxa are presented in from Forey [12]. Character 109 ‘‘ventral keel scales absent (0), Information S5. present (1)’’, was proposed by Friedman & Coates [13]. 2.3 Phylogenetic results. The topology of the strict consen- Furthermore, we propose one new character: character 110 sus tree (Figure 22) obtained from the 22 shortest trees (length ‘‘ventral swelling of the palatoquadrate absent (0), present (1)’’ = 287; consistency index = 0.4146; retention index = 0.6929) (Figure 20). places Megalocoelacanthus as the sister-taxon of Libys within clade 16. Finally, the data matrix includes the coelacanths recently Three unambiguous synapomorphies support the node [Mega- described: Piveteauia [32], Swenzia [8], Holopterygius [13], Parnaibaia locoelacanthus + Libys]: a supraorbital sensory line canal opening [14], Guizhoucoelacanthus [9], Rebellatrix [28], and several taxa through a large and continuous groove (23[3], that is a non- (Axelia, Euporosteus, Indocoelacanthus, Lualabaea, Ticinepomis, and homoplastic synapomorphy), the infraorbital, jugal, and preoper- Wimania) that were coded by Forey [12] but excluded from his cular sensory line canals opening through a large and continuous final analysis because of their high amount of missing data or the groove crossed by pillars (50[2]), and a robust prearticular and instability they were raising in the topology. principal coronoid, marked with fine striations (68[1]). Addition- 2.2 Searching methods. The data matrix (Information S2) ally, clade 16 is supported by six ambiguous synapomorphies: the was constructed in Mesquite 2.74 [33]. It comprises 39 taxa and presence of snout bones consolidated (2[1]), the presence of a 110 anatomical characters including 88 cranial and 22 postcranial preoperculum developed as a posterior tube-like canal-bearing anatomical characters (Information S1). Maximum parsimony portion and an anterior blade-like portion (39[1]), the absence of analyses were carried out using the software PAUP 4.0b10 [34]. A ornamentation upon cheek bones (49[0]), the optic foramen lying

PLOS ONE | www.plosone.org 19 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 18. Megalocoelacanthus dobiei Schwimmer, Stewart & Williams, 1994, AMNH FF 20267 from lower Campanian of the Niobrara Formation. Shoulder girdle. A, Left cleithrum in lateral (left) and medial (right) views. B, right cleithrum in lateral (left) and medial (right) views. C, clavicle in posterior (top) and anterior (bottom) views. D, right scapulocoracoid in medial (top) and lateral (bottom) views. E, left scapulocoracoid in medial (top) and lateral (bottom) views. Abbreviations: Cl, cleithrum; Ecl, extracleithrum. Scale bar = 10 cm (A–B), 5 cm (C–E). doi:10.1371/journal.pone.0049911.g018 within an interorbital ossification or cartilage separate from the different topologies where Libys was alternatively the sister-group basisphenoid (70[1]), the presence of a forked anocleithrum of the clade [Diplurus [Chinlea [Mawsonia + Axelrodichthys]]], the (89[1]). The ambiguity arises from the fact that characters 2 and sister-group of the clade [Macropoma [Swenzia + Latimeria]], or the 70 are known in Megalocoelacanthus but not in Libys, and that sister-group of Garnbergia. The last two hypotheses are corroborat- characters 39, 49 and 70 are known in Libys but not in ed in our topology, suggesting that Libys is more closely related to Megalocoelacanthus. Depending on the optimization the state of Latimeria than to Mawsonia. However, these last two topologies each character cited can be a synapomorphy of clade 16 (FAST suggested that Latimeria is more closely related to Libys than it is to optimization) or an autapomorphy of the terminal taxa in which it both Holophagus and Undina. This is inconsistent with our results is known (SLOW optimization). and those of Geng et al. [9] where Libys is the sister group of a The question of the phylogenetic affinities of M. dobei with other clade including Holophagus, Undina, Macropoma, Swenzia and coelacanths sheds a new light on the phylogenetic position of its Latimeria. sister-taxon. Forey [12] supported a sister-group relationship with Our analysis provides new insights in the interrelationships of the clade [Diplurus + Mawsonia] based on a single homoplastic Latimerioidei. It provides new information on the unsolved synapomorphy: the reduction of the ornament (49[0]). The relationships between taxa that are well informed, and on the position of Libys was undetermined within Latimerioidei in affinities of taxa that were traditionally considered as problematic, Cle´ment [8] and Friedman & Coates [13]. Cle´ment [8] presented

PLOS ONE | www.plosone.org 20 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 20. Comparison of the palatoquadrate of two actinis- tians showing the presence/absence of the ventral swelling of the pterygoid (arrow) coded as character 110. A, Axelrodichthys, right palatoquadrate in medial view (modified from Maisey 1986). B, Latimeria, left palatoquadrate in medial view (modified from Forey 1998). Scale bar = 20 mm. doi:10.1371/journal.pone.0049911.g020

Figure 19. Megalocoelacanthus dobiei Schwimmer, Stewart & Taxa that have been previously considered as mawsoniids Williams, 1994. A, neural spine of the holotype specimen CCK 88-2-1 (Diplurus, Chinlea, Mawsonia, Axelrodichthys, Parnaibaia) are here from lower Campanian of the Blufftown Formation in lateral (left) and anterior (right) views. B, scale of AMNH FF 20267 from lower retained within clade 21. This clade is supported by four Campanian of the Niobrara Formation. Scale bar = 5 cm. unambiguous synapomorphies: the absence of a supratemporal doi:10.1371/journal.pone.0049911.g019 descending process (14[0]), the presence of an unmodified posterior end of the coronoid (56[0]), the presence of ossified ribs and excluded from the analysis because of the presence of many (92[1]), a feature unique to this clade, and the presence of missing data. differentiated scale ornaments (104[1]). The absence of the The position of Garnbergia was unstable in previous analyses: it supratemporal descending process is interpreted as a secondary was alternatively placed within Mawsoniidae [8,12], as the sister- loss because its presence is here a synapomorphy of clade 3 (FAST group of Libys within Latimeriidae [8], or as the sister group of the optimization) or clade 5 (SLOW optimization). The unambiguous least inclusive clade containing Latimeria, Mawsonia and Coelacanthus synapomorphies that support clade 21 were also supporting the [10]. In contrast, sister-group relationship between Garnbergia and clade recognised as Mawsoniidae (that is [Diplurus [[Mawsonia + Latimerioidei is suggested by the present analysis. This hypothesis Axelrodichthys] + [Chinlea + Parnaibaia]]]) in the unconstrained is supported by a single unambiguous synapomorphy: the presence analysis of Yabumoto [14]. Additional synapomorphies included of 8 to 9 fin rays on the first dorsal fin (96[1]). This synapomorphy the loss of the suboperculum (32[0]) and of the subopercular is homoplastic and also represents an apomorphy of Laugia. Within branch of the mandibular sensory line canal (60[0]). However, clade 13, this character is well documented and in all taxa where it these reversions are actually symplesiomorphies that were is known (except Mawsonia, Megalocoelacanthus and Swenzia) possess 8 misinterpreted because of the lack of resolution in other to 9 fin rays on the first dorsal fin. Nevertheless, this character latimerioids relationships. Indeed, other latimerioids that have a should be taken with caution: fine rays can be easily lost after the suboperculum (Holophagus, Libys, and Latimeria) and a subopercular death of the animal, and the number of fine rays is variable branch of the mandibular sensory line canal (Holophagus, Latimeria, Libys Macropoma Megalocoelacanthus between individuals in the extant coelacanth Latimeria. , and here ) formed a polytomy with Mawsoniidae in Yabumoto’s (2008) results. Consequently, Clade 13 is supported by a single unambiguous synapomorphy: the most parsimonious scenario in the phylogenetic analysis of the presence of denticles on the fin rays of the D1 (98[1]). This Yabumoto [14] was favouring a reversion in Mawsoniidae instead clade was recognized as Latimerioidei in previous studies. The of the retention of the plesiomorphic condition in this clade, while presence of a postparietal descending process (13[1]) that was Holophagus, Libys, Macropoma, and Latimeria would have acquired previously considered as a synapomorphy of this clade [9,12] is convergently the apomorphic condition. Thanks to the better here an ambiguous synapomorphy of clade 13. The ambiguity resolution the relationships of these taxa, these putative synapo- comes from the lack of information for this character in Garnbergia morphies should be there considered as symplesiomorphies. and Rebellatrix the most closely related taxa of latimerioids. Within clade 21, the sister-group relationship between Mawsonia Consequently, the presence of the descending process of the and Axelrodichthys is retained, and strongly supported by six postparietal could be a synapomorphy of clade 11 [Rebellatrix unambiguous synapomorphies, including three non-homoplastic [Garnbergia [Latimerioidei]]] under FAST optimization or a ones: the extrascapulars forming part of the skull roof (16[1]), the synapomorphy of clade 13 (Latimerioidei) under SLOW optimi- presence of a ventral process on the postorbital (41[1]), and the zation. principal coronoid sutured to the angular (66[1]). The relation-

PLOS ONE | www.plosone.org 21 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Parnaibaia as the sister-group of [Mawsonia + Axelrodichthys]. Our strict consensus tree is inconsistent with these hypotheses and rather supports the sister-group relationship of Chinlea and [Mawsonia + Axelrodichthys] based on three homoplastic synapo- morphies: the presence of coarse rugosities on the parietals and postparietals (27[2]), the presence of coarse rugosity on cheek bones (49[2]) and rugose scales (106[1]). Ticinepomis branches at the base of clade 14 (Figure 22). Its position changed dramatically compared to Cloutier’s hypothesis (clade F in [10]) in which it was the sister-group of a clade including Wimania, Axelia, and Coelacanthus. This group appears here to be polyphyletic. Ticinepomis was subsequently excluded by Forey [12] from the analysis because it was raising instability in the intrarelationships of the sister-group of the clade [Coccoderma + Laugia]. As pointed out by Forey [12], the irresolution resulting from the inclusion of Ticinepomis arose from the contradiction in the distribution of its character states rather than to the lack of data. The position of Ticinepomis is here supported by a single unambiguous synapomorphy: the presence of expanded median fin rays (103[1]). The same position was found in the strict consensus tree of the first analysis we performed (Figure 21), but the node was only supported by ambiguous synapomorphies. The presence of expanded median fin rays on the D1 (103[1]) is only found in Ticinepomis, Libys, and Holophagus. Undina, Macropoma, and Latimeria display unexpanded median fin rays (103[0]) and this condition is a synapomorphy of clade 18 interpreted as a reversal toward the ancestral condition of actinistians. Additionally, clade 14 is supported by six ambiguous synapo- morphies: the presence of a several median rostrals (3[1]), the presence of an anterior branch of the supratemporal commissure (22[1]), the absence of a spiracular (30[0]), the presence of a subopercular branch of the mandibular sensory line canal (60[1]), the presence of an ascending lamina of the parasphenoid (79[1]), and the presence of a ventral swelling of the palatoquadrate (110[1]). The snout is generally a poorly preserved region of the skull in fossil actinistians [12]. Among Latimerioidei, it is only observed in Axelrodichthys and Diplurus, which possess a single rostral (3[0]), and in Macropoma, Latimeria and Parnaibaia which possess several median rostrals (3[1]). This condition cannot be assessed for other genera of the clade. The presence of several rostrals (3[1]) is plesio- morphic in actinistians, and lost at most in clade 3, the least inclusive clade containing Hadronector, Lochmocercus and Latimeria. Depending on character optimization, reversion toward the ancestral state occurs independently in Parnaibaia and Latimeria (SLOW optimization), or in Parnaibaia and clade 14 (FAST optimization). The latter hypothesis would support Forey’s analysis [12] where the presence of several rostrals is a synapomorphy of Figure 21. Result of the first phylogenetic analysis based on 39 Megalocoelacanthus taxa and 110 characters. Strict consensus tree of the 584 equally Latimeriidae. Then, it would also suggest that parsimonious trees (length = 288; consistency index = 0.4132; retention possessed such a condition. Re-examination of Libys specimens, index = 0.6938). Branch in grey is supported only by ambiguous and better-preserved material of Ticinepomis and other latimeriids synapomorphies. are needed to better understand the polarity of this character. doi:10.1371/journal.pone.0049911.g021 Among actinistians, the presence of an anterior branch of the supratemporal commissure (22[1]) over the postparietal is only ships between Lualabaea, Parnaibaia, Chinlea and the clade [Mawsonia observed in Macropoma, Swenzia, and Latimeria. The ambiguity of + Axelrodichthys] were unresolved in the strict consensus tree from the polarity of this character arises from the fact that all other the first analysis (Figure 21) due to the instability of Lualabaea genera belonging to clade 14 are scored with a question mark for (Figure 21B). The first strict consensus tree proposed by Yabumoto this character. FAST optimization supposes that the presence of (2008) supported a sister-group relationship between Parnabaia and the supratemporal commissure is a synapomorphy of clade 14, Chinlea based on the presence of two homoplastic synapomorphies, whereas SLOW optimization supports it as a synapomorphy of the presence of anterior and posterior parietals of similar size clade 19 [Macropoma [Latimeria + Swenzia]]. This character is (8[0]), and the presence of an angle on the anterior end of the unknown in Undina and Ticinepomis [12,37,38]. In Megalocoela- lachrymojugal (36[1]). Otherwise, its second consensus tree canthus, the posterior portion of the postparietal is covered by resulting from successive weighting procedure was supporting grooves directed anteriorly (Figures 5, 7), but the state of

PLOS ONE | www.plosone.org 22 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

Figure 22. Result of the second phylogenetic analysis based on 37 taxa and 110 characters. Strict consensus tree of the 22 shortest trees (length = 287; consistency index = 0.4146; retention index = 0.6929). Nodes are numbered from 1 to 30, and the list of apomorphies for each node and terminal taxon is given in Information S5. Numbers on the left of the node indicate the Bremer decay indices. Bootstrap values are indicated after the Bremer decay indices if superior to 50%. doi:10.1371/journal.pone.0049911.g022 preservation of the skull roof does not enable to clearly identify according to our analysis. There is a subopercular branch of the them as anterior branches of the supratemporal commissure. In its mandibular sensory line canal (60[1]) in Megalocoelacanthus, Libys, sister-group Libys polypterus, the supratemporal commissure does Holophagus, Macropoma, and Latimeria. This condition is thus a non- not present anterior branches (figure 3.17 in [12]). The presence of homoplastic synapomorphy of at least clade 15 (SLOW optimi- such a feature has been considered in Holophagus gulo based on the zation), and is therefore inferred in Undina and Swenzia. Potentially, presence of longitudinal grooves extending from the extrascapulars it could be shared by Ticinepomis (FAST optimization) but the (figure 3.18 in [12]). If the presence of anterior branches of the available material is too poorly preserved to assess its condition supratemporal commissure is confirmed in Holophagus, our [38]. The ascending lamina of the parasphenoid is originally topology would imply that this feature could be a synapomorphy absent in actinistians (79[0]). Its presence (79[1]) is observed in of clade 17. Megalocoelacanthus, Undina, Macropoma, Swenzia, and Latimeria, and is Characters (30), (60), (79), and (110) have an ambiguous polarity therefore a non-homoplastic synapomorphy of at least clade 15 within clade 14 caused by the lack of data regarding Ticinepomis. (SLOW optimization). The presence of the ascending lamina of The presence of a spiracular (30[1]), for instance, is plesiomorphic the parasphenoid is inferred in Ticinepomis under FAST optimiza- for actinistians. The loss of the spiracular (30[0]) is convergent in tion. The ventral swelling of the palatoquadrate (110) is here a clade 28 [Laugia + Coccoderma], in clade 24 [Mawsonia + newly recognised character. It is only observed in Holophagus, Axelrodichthys], and in clade 14. A reversion to the ancestral state Latimeria, Libys, Macropoma, Megalocoelacanthus, and Undina, and (30[1]) occurred in clade 20 [Latimeria + Swenzia]. The spiracular is unknown in Ticinepomis and Swenzia. Like previous characters, the absent in Libys and this condition is inferred for Megalocoelacanthus presence of this ventrally swelling is either a synapomorphy of

PLOS ONE | www.plosone.org 23 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus clade 15 (SLOW optimization) or of clade 14 (FAST optimiza- synapomorphy’’ [42,43,45,46]. Apomorphy-based definitions are tion). not used here because they encounter three problems: character Clade 15 [[Megalocoelacanthus + Libys] + [Holophagus + Undina + ambiguity, variation in characters optimization, and homoplasy [Macropoma [Latimeria + Swenzia]]]] is supported by a two [46]. unambiguous synapomorphies: the presence of a hook-shaped A node-based definition specifies the membership of a taxon by dentary (57[1]), and the presence of multiple opening for the ‘‘the least inclusive clade that contains at least two internal lateral line on scales (105[1]). the presence of a hook-shaped specifiers’’, while a stem-based definition specifies the membership dentary (57[1]) was previously considered as a non-homoplastic of a taxon by ‘‘the most inclusive clade that contains at least one synapomorphy of the clade [Whiteia + Latimerioidei] in the internal specifier’’ [46]. Stem-based and node-based definitions constrained analysis of Forey [12]. This is not consistent with our thus rely on the use of a definitional component termed as analysis in which this change is convergent for clades 15, 22, and specifiers, i.e species or specimens stated in the phylogenetic in Whiteia. As noted before, the relationships between Indocoela- definition as a reference point. Both definitional types have at least canthus, Undina, Holophagus, and the clade [Macropoma [Latimeria + one internal specifier (anchor within the ingroup defined), but Swenzia]] were unresolved in the strict consensus tree of first stem-based definition contains optionally an external specifier to analysis (Figure 21) because of the unstable position of Indocoela- define the exclusion group. An optional definitional component canthus due to its incompleteness. Undina is more closely related to termed as ‘‘qualifier’’ can also be used to provide conditions on the [Macropoma [Latimeria + Swenzia]] than Holophagus based on two clade membership. Qualifiers can be species, specimens, or unambiguous synapomorphies, the presence of the oral pit line features of species or specimens. removed from the center of ossification (59[1]), and the presence We propose here phylogenetic definitions to the taxa Latimer- of unexpanded median fin rays (103[0]). This is inconsistent with iidae, Mawsoniidae and Latimerioidei. The genera Latimeria Smith previous hypotheses [9,12,14] that were suggesting a sister-group 1938 and Mawsonia Woodward 1907 are given as reference taxa relationship between Holophagus and Undina. Within clade 15, the within Latimerioidei because they are well-known and complete, sister-group relationship between the clade [Latimeria + Swenzia] deeply nested, and their position is stable in the successive and Macropoma is retained in both analyses we performed topologies that have been proposed up to now [8–14,28]. We thus (Figures 21, 22). This sister-group relationship is supported by consider Latimeria chalumnae and Mawsonia gigas reliable enough to three unambiguous synapomorphies: the presence of a preoper- preserve the taxonomic content of the taxa we define here. culum developed as a posterior tube-like canal-bearing portion Our study supports the presence of two major clades of and an anterior blade-like portion (39[1]), the presence of an coelacanths (clade 14 and 21) within a more inclusive clade (clade anterodorsal excavation in the postorbital (40[1]), that is a feature 13) that are retained in the strict consensus trees of both analyses exclusive to this clade, and the presence of less than eight fin rays performed (Figures 21, 22). However, their inter- and intra- on the first dorsal fine (96[2]). Additionally, this clade is supported relationships are still weakly supported. Indeed clade 14 is by two ambiguous synapomorphies. The presence of snout bones supported by a single unambiguous synapomorphy in the second consolidated (2[1]) is a synapomorphy of clade 19 under FAST analysis (Figure 22), and only by ambiguous synapomorphies in optimization that is subsequently lost in Latimeria,orisa the first analysis we performed (Figure 21), raising a polytomy convergence in Macropoma and Swenzia under SLOW optimization. between Ticinepomis, clade 21 and clade 15 when branches with The presence of a splenial without ornament (64[0]) is alterna- minimum length of zero are collapsed. Consequently, we decided tively a synapomorphy of clade 19 (FAST optimization), or a to give a stem-based definition of Latimeriidae and Mawsoniidae. synapomorphy of clade 17 that is lost by convergence in Holophagus Moreover, it is worth noting that the family Mawsoniidae as and Undina. The sister-taxon relationship between Latimeria and coined by Schultze [15] is a nomen nudum. Actually, the name is Swenzia is supported by two unambiguous synapomorphies: the invalid in regard with the Article 13.1 of the ICZN [48], because absence of pit lines making postparietals (26[1]), and the presence neither diagnostic characters nor definitions are stated in the of a spiracular (30[1]). These synapomorphies where also publication. A diagnosis was subsequently given by Forey [12], but supporting this clade in the phylogenetic analysis of Cle´ment [8] no definition has been stated since the name was coined. Based on and both represent a reversal towards the ancestral condition in previous topology and ours, we propose the following definitions: actinistians. Latimeriidae Berg, 1940: the most inclusive clade containing Latimeria chalumnae Smith, 1938 but excluding Mawsonia gigas 3. Systematic paleontology Woodward, 1907. Based on the topology proposed by Cloutier [10], Schultze [15] Mawsoniidae Schultze, 1993: the most inclusive clade contain- erected the suborder Latimerioidei (node 15 in [10]) comprising ing Mawsonia gigas Woodward, 1907 but excluding Latimeria the family Mawsoniidae Schultze, 1993 (node I in [10]) and the chalumnae Smith, 1938. family Latimeriidae Berg, 1940 (node 16 in [10]). Although The dichotomy between Latimeriidae and Mawsoniidae is diagnoses for each taxa cited above were given by Berg [39] and traditionally recognized in the phylogenies, stating the relationship Forey [12], these taxa remain undefined. The confusion between ‘‘Latimerioidei = Latimeriidae + Mawsoniidae’’. In order to taxon definition and taxon diagnosis has been discussed by handle the taxa Latimerioidei in the case of relocation of species, Ghiselin [40] and Rowe [41], who emphasized the difference clades, or changes in the apomorphies, we use a node-stem triplet between these two expressions. (NST). The NST is the only phylogenetic definition that preserves Phylogenetic taxonomy aims to formulate taxonomic definitions the taxonomic content at a dichotomy by combining a node-based based on the phylogenetic relationships and to make a clear definition of a taxon with two stem-based definitions of more distinction between taxonomic diagnosis and taxonomic definition inclusive taxa [49]. Criteria of diversity, morphology and tradition [42–46]. Three classes of definitions have been initially stated in are recommended for establishing a NST for a certain taxa [49]. phylogenetic taxonomy: apomorphy-based, stem-based, and node- With the NST the traditional equivalence statement ‘‘Latimer- based definitions [42,47]. Apomorphy-based definition makes ioidei = Latimeriidae + Mawsoniidae’’ is anchored and will reference to characters, and defines the membership of a taxon as always be stable in the case of a taxon branching basally to the the clade stemming from the ‘‘first ancestor with a particular clade 14 or to clade 21. The NTS we propose consists in a node-

PLOS ONE | www.plosone.org 24 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus based definition for Latimerioidei composed of two stem-based Discussion and Conclusions taxa, Latimeriidae and Mawsoniidae. Using Latimeria chalumnae and Mawsonia gigas as nested specifiers, the phylogenetic definition Since its first description, Megalocoelacanthus was previously of Latimerioidei is proposed as follows: related to the latimeriids Latimeria and Macropoma based on few Latimerioidei Schultze, 1993: the least inclusive clade contain- meristic data [1]. Our phylogenetic analysis of Megalocoelacanthus ing Latimeria chalumnae Smith, 1938 and Mawsonia gigas Woodward, supports the sister-group relationship with Libys, an Upper Jurassic 1907; genus from Bavaria, Germany. Although it is significantly smaller Latimeriidae Berg, 1940: the most inclusive clade containing in size, Libys shares many features with Megalocoelacanthus. Notably, Latimeria chalumnae Smith, 1938 but excluding Mawsonia gigas this genus also possesses a supraorbital sensory line canal that Woodward, 1907; opens through a large, continuous groove stretched by pillars on Mawsoniidae Schultze, 1993: the most inclusive clade con- either side of the parietonasal shield. When specimens of Libys are tinaining Mawsonia gigas Woodward, 1907 but excluding Latimeria observed under binocular microscope (Dutel pers. obs. on BSPG chalumnae Smith, 1938. XIV 501b, BSPG 1870 XIV 502, BMNH P3337), one can Megalocoelacanthus was previously included in the family Coela- distinguish a suture between the base of adjacent pillars. It is thus canthidae Agassiz, 1844 by Schwimmer et al. [1]. However, the probable that the same condition is present in Megalocoelacanthus, family Coelacanthidae as defined by Agassiz [50] should not be but because of the poor preservation, the suture between the longer recognized. It is actually inconsistent with the topology supraorbitals cannot be seen. Despite the size difference, the lower obtained here and in previous studies because it does not represent jaw of Megalocoelacanthus and Libys are virtually identical, i.e. a a monophyletic group [8–14,28]. Based on the hypothesis of slender and elongated mandible opened by large pores for the relatedness obtained here and using the new definitions we stated, mandibular sensory line canal. These two genera also share the we thus propose the following taxonomy for Megalocoelacanthus as absence of a suture between the parasphenoid and the basisphe- well as a new diagnosis: noid, which suggests fusion of the two bones, the presence of a OSTEICHTHYES Huxley, 1880 [51]. narrow and unornamented parietonasal shield, and the palato- SARCOPTERYGII Romer, 1955 [52]. quadrate being very deep and short in length with a ventral ACTINISTIA Cope, 1891 [53]. swelling on the palate. Very few elements from the postcranial LATIMERIOIDEI Schultze, 1993 [15] new definition. skeleton are present on the specimen AMNH FF 20267. However, LATIMERIIDAE Berg, 1940 [49] new definition. the pectoral girdle which is preserved is virtually identical in shape Megalocoelacanthus dobiei Schwimmer, Stewart & Wil- and proportion in Megalocoelacanthus and Libys: the cleithrum is liams, 1994. slender and elongated, and the clavicle is relatively very small Holotype. CCK-8-2-1, basisphenoid, left lower jaw, right and compared to the whole pectoral girdle. left palatoquadrates, pectoral girdles, left opercular, zygal plate, A very interesting aspect of Megalocoelacanthus lies in the shape many branchial elements, dorsal fin spine, many indeterminate and the proportion of its skull. In genera such as Axelrodichthys and bones. Mawsonia where the skull is shallow and elongated, the basisphe- Paratype. AUMP 3834, right mandible, right principal noid has a short dorsum sellae, with laterally well-expanded wings, coronoid, right and left palatoquadrates and isolated metapter- as well as a palatoquadrate longer than high [5]. In contrast, the ygoids, right and left autopalatines, right gular plate, left opercular, well preserved basisphenoid of the holotype specimen CCK 88-2-1 right certohyal, single indeterminate branchial. is deeper and narrower than in these genera. The palatoquadrate Referred material. FMNH P27534, palatoquadrate; CCK of Megalocoelacanthus is deeper than long, and appears to be 93-6-1, and AUMP 3944, distal quadrate fragments; CCK 93-13- proportionally shorter in length than those of the latimeriids 1, right angular fragment; AMNH 6643, principal coronoid Latimeria, Macropoma, Holophagus and of the mawsoniids Mawsonia, fragment; AMNH FF 20267, ethmosphenoid and otoccipital Axelrodichthys and Parnaibaia. The shoulder girdle also appears to be portion of the skull, isolated snout, right and left lower jaws, much deeper and more slender than that of the mawsoniids isolated right and left articulars, right and left palatoquadrates, Diplurus, Mawsonia and Axelrodichthys. Megalocoelacanthus clearly right and left autopalatines, right and left gular plates, urohyal, shows features of Latimeriidae, but the proportions of the elements basibranchial associated with tooth plates, undetermined branchial of the skull and shoulder girdle are closer to what can be observed arches, ceratohyal, left symplectic, right and left operculars, in Libys than in other genera of this clade. shoulder girdle (right and left cleithrums, clavicles, scapulocor- In all the latimerioids examined, the parietals are much wider acoids), isolated indeterminate elements. than the supraorbitals and the parietonasal shield thus represents Diagnosis (revised). Parietonasal shield narrow, and longer most of the width of skull roof. Megalocoelacanthus and Libys are also than the posparietal shield. Supraorbital sensory line canal very unique in that their parietonasal shield is considerably opening through a large groove crossed by slender pillars. Ventral narrower compared to other coelacanths, and the skull is mainly descending processes present on the supratemporal and on the roofed by the supraorbitals that are lying alongside the parietal. Basisphenoid very deep dorsoventrally. Palatoquadrate parietonasal shield. However, we cannot determine whether the deeper than long, with distinct swelling extending from the ventral bony surface within the vacuities of the supraorbital sensory line pterygoid margin immediately anterior to the quadrate. Mandibles canal is a lateral extension of the parietonasal shield that is relatively elongate posterior to the articular; articular sutured to overlapped by the supratemporal. Although the skull of AMNH the angular. Medial surfaces of the prearticular, palatoquadrate, FF 20267 is strongly flattened laterally, the skull roof was most and coronoid covered with tubercular shagreen. No marginal probably well-vaulted. In any case, it is clear that the top of the teeth on the mandible. Coronoid large, with subcircular ventral skull was much narrower than the buccal floor. When the jaws and margin. Lateral surface of angular bears large pores of the sensory the gular plates of AMNH FF 20267 are assembled, the minimum line canal and very faint longitudinal grooves on its posterior width of the buccal floor is much wider than the skull roof width. portion. Gular and operculum lack external ornamentation. Taken together, these elements enable us to depict Megalocoela- Operculum subrhomboidal, with sharply angled anteroventral canthus as a large-sized coelacanth with a short, laterally margin. Gular plates diverge strongly along posterior midline. compressed and very deep skull, with a bell-like shape in transverse

PLOS ONE | www.plosone.org 25 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus section. This is thus quite different from the more ovoid transverse poorly known genus Trachymetopon from the lower Toarcian of the section of the skull of the extant coelacanth Latimeria chalumnae. Posidonia Shale of Germany [59]. Mawsonia was certainly non- Coelacanths have been depicted as a conservative group that marine and restricted to continental and estuarine environments experienced little anatomical change during their evolution. [2,60], whereas Trachymetopon was clearly marine. Although the However, this widespread idea has been challenged at several phylogenetic position of Trachymetopon has still to be elucidated, our times by the discovery of Paleozoic coelacanths such as Allenypterus study suggests that large coelacanths evolved in at least two [17,54], Holopterygius [13] and Miguashaia [21], whose morpholo- different lineages during the Mesozoic, in both marine and non- gies differ significantly from that of Latimeria. It now appears that marine environments. coelacanths experienced a wide range of morphologies and ecologies very early in their evolutionary history. Based on Supporting Information geometric morphometric analysis Friedman & Coates [13] showed that the highest morphological disparity in coelacanths was Information S1 Character list. reached by the Middle Devonian, but dropped in post-Carbon- (DOC) iferous forms despite a significant increase in the taxonomic Information S2 Data matrix used in the phylogenetic diversity. Indeed, Mesozoic coelacanths actually fit to the analysis. ‘‘Latimeria bodyplan’’ and anatomical variations in Mesozoic (DOC) coelacanths seem to lie in variations of proportion of the skeletal elements rather than in radical morphological shifts. Information S3 Nexus file of the first phylogenetic The evolution of Mesozoic coelacanths also appears to now be analysis. marked by a significant increase in body size in at least two (TXT) lineages during the Cretaceous. Nevertheless, the skull morphol- Information S4 Nexus file of the second phylogenetic ogy of these forms is far from being homogenous and clearly falls analysis. into two morphotypes: long, shallow, wide skulls in the mawsoniids (TXT) Mawsonia and Axelrodichthys and short, deep, narrow skulls in the Information S5 Diagnostic information for the node and latimeriids such as in Latimeria, Macropoma, which is amplified in terminal taxa of the strict consensus tree illustrated Libys and Megalocoelacanthus. It is worth noting that these variations Figure 22. may focus considerable interest in the intracranial joint kinetic, (DOC) and thereby in the feeding strategy of these large-sized taxa. Based on a better understanding of the intracranial joint kinetic in Latimeria, this question will deserve further investigation in the Acknowledgments future. The AMNH specimen was generously donated by Robert G. Goelet, The paleoenvironments represented by the occurrences of Chairman Emeritus of the Board of Trustees at the American Museum of Megalocoelacanthus differ widely. The marine chalks of western Natural History. Anthony Maltese and Mike Triebold (Rocky Mountain Kansas and western Alabama represent offshore, deep-water Dinosaur Resource Center, Woodland Park), who collected and prepared marine environments, whereas the detrital sediments from eastern the material, are thanked for their useful information about the geological Alabama, Georgia and New Jersey (including the holotype context of the Kansas specimen. Julien Massoni (Universite´ Paris-Sud XI, Orsay), Jocelyn Falconnet (Paris), Alan Pradel (AMNH, New-York City) occurrence) represent near shore, shallow marginal marine to and Damien Germain (MNHN, Paris) are thanked for their useful advice estuarine environments. Given the range of occurrences, this and comments on the manuscript, and their help in the use of the suggests that Megalocoelacanthus was eurytopic and favored both phylogeny software. Sophie Fernandez and Pascal Leroch’ (MNHN, Paris) marine and brackish waters. The records of Megalocoelacanthus [55] are thanked for their advice and help in making the illustrations. We thank collectively suggest that it was a fairly common fish, but it is not Lorraine Meeker and Chester Tarka (AMNH, New York), who took the frequently recognized in fossil assemblages and its distribution may photographs of the AMNH specimen. We also thank the curators and still be underestimated. Further study will have to be carried on in collection managers Markus Moser (BSPG, Munich) and Zerina Johanson (BMNH, London) for permitting the examination of the coelacanth order to determine the potential ecological niches that these material under their care, and for the warm welcome in their respective common large, toothless, coelacanths were occupying in the institutions. Daphne Soares and the two other anonymous reviewers of this Western Interior Seaway during the Late Cretaceous. article are thanked for their useful comments, which improved the quality By comparison with Libys and Latimeria, the length of AMNH FF of the manuscript. This work is a contribution to the ANR TERRES 20267 is estimated to range between 2.30 m and 3 m. However, a Programme (ANR-2010-BLAN-607-03). large isolated principal coronoid found near the holotype site in eastern Alabama extrapolates the maximum length of Megalocoe- Author Contributions , lacanthus to 4.5 m [56]. Similar dimensions were previously Conceived and designed the experiments: GC JGM PJ. Performed the known only through the genus Mawsonia from the Lower experiments: HD GC. Analyzed the data: HD GC DS JGM. Contributed Cretaceous of Brazil, Morocco and Niger, which largest specimens reagents/materials/analysis tools: HD GC DS JGM PJ MH. Wrote the are estimated to range between 3.5 and 6.3 m [2,57,58], and the paper: HD GC DS JGM PJ MH. Scientific illustrations: HD.

References 1. Schwimmer DR, Stewart JD, Williams GD (1994) Giant fossil coelacanths of the 4.GalloV,CarvalhodeMSS,SantosHRS(2010)Newoccurenceof Late Cretaceous in the eastern United States. Geology 22: 503–506. {Mawsoniidae (Sarcopterygii, Actinistia) in the Morro do Chaves Formation, 2. Carvalho de MSS, Maisey JG (2008) New occurrence of Mawsonia (Sarcopter- Lower Cretaceous of the Sergipe-Alagoas Basin, Northeastern Brazil. Boletim do ygii: Actinistia) from the Early Cretaceous of the Sanfranciscana Basin, Minas Museu Paraense Emilio Goeldi 5: 195–205. Gerais, southeastern Brazil. Geological Society, London, Special Publications 5. Maisey JG (1986) Coelacanths from the Lower Cretaceous of Brazil. American 295: 109–144. Museum Novitates 2866: 1–30. 3. Cavin L, Forey PL (2004) New mawsoniid coelacanth (Sarcopterygii: Actinistia) 6. Soto M, Carvalho de MSS, Maisey JG, Perea D, Da Silva J (2012) Coelacanth remains from the Cretaceous of the Kem Kem beds, Southern Morocco. In: remains from the late Jurassic-?earliest Cretaceous of Uruguay: the southern- Arratia G, Tintori A, editors. Mesozoic Fishes 3 – Systematics, Paleoenviron- most occurence of the Mawsoniidae. Journal of Vertebrate Paleontology 32: ments and Biodiversity. Mu¨nchen: Verlag Dr. Friedrich Pfeil. 493–506. 350–537.

PLOS ONE | www.plosone.org 26 November 2012 | Volume 7 | Issue 11 | e49911 The Giant Cretaceous Coelacanth Megalocoelacanthus

7. Wenz S (1981) Un cœlacanthe ge´ant, Mawsonia lavocati Tabaste, de l’Albien-base 33. Maddison WP, Maddison DR (2010) Mesquite: a modular system for du Ce´nomanien du sud marocain. Annales de Pale´ontologie (Verte´bre´s) 67: 1– evolutionary analysis. Version 2.74. http://mesquiteproject.org. 20. 34. Swofford DL (2002) PAUP*. Phylogenetic analysis using parcimony (*and other 8. Cle´ment G (2005) A new coelacanth (Actinistia, Sarcopterygii) from the Jurassic methods). Version 4. Sunderland, Massachusetts: Sinauer Associates. of France, and the question of the closest relative fossil to Latimeria. Journal of 35. Sorenson MD, Franzosa EA (2007) TreeRot. Version 3. 3 ed. Boston: Boston Vertebrate Paleontology 25: 481–491. University. 9. Geng B-H, Zhu M, Jin F (2009) A revision and phylogenetic analysis of 36. Nixon KC (2002) WinClada. Version 1.00.08. New-York: Ithaca. Guizhoucoelacanthus (Sarcopterygii, Actinistia) from the Triassic of China. 37. Jain SL (1974) Indocoelacanthus robustus n. gen., n. sp. (Coelacanthidae, Lower Vertebrata PalAsiatica 47: 311–329. Jurassic), the first fossil coelacanth from India. Journal of Paleontology 48: 49– 10. Cloutier R (1991) Patterns, trends, and rates of evolution within the Actinistia. 62. Environmental Biology of Fishes 32: 23–58. 38. Rieppel O (1980) A new coelacanth from the Middle Triassic of Monte San 11. Forey PL (1991) Latimeria chalumnae and its pedigree. Environmental Biology of Giorgio, Switzerland. Eclogae Geologicae Helvetiae, Basel 73: 921–939. Fishes 32: 75–97. 39. Berg LS (1940) Classification of fishes, both recent and fossil. Trudy 12. Forey PL (1998) History of the coelacanth fishes. London: Chapman and Hall Zoologicheskogo Instituta, Akademiya Nauk SSSR, Leningrad 5: 1–141. 440 p. 40. Ghiselin MT (1984) ‘‘Definition,’’ ‘‘character,’’ and other equivocal terms. 13. Friedman M, Coates MI (2006) A newly recognized fossil coelacanth highlights the early morphological diversification of the clade. Proceedings of the Royal Systematic Zoology 33: 104–110. Society B: Biological Sciences 273: 245–250. 41. Rowe T (1987) Definition and diagnosis in the phylogenetic system. Systematic 14. Yabumoto Y (2008) A new Mesozoic coelacanth from Brazil (Sarcopterygii, Zoology 36: 208–211. Actinistia). Paleontological Research 12: 329–343. 42. Queiroz de K, Gauthier J (1990) Phylogeny as a Central Principle in Taxonomy: 15. Schultze H-P (1993) Osteichthyes: Sarcopterygii. In: Benton MJ, editor. The Phylogenetic Definitions of Taxon Names. Systematic Biology 39: 307–322. Fossil Record 2. London: Chapman & Hall. 657–663. 43. Queiroz de K, Gauthier J (1992) Phylogenetic taxonomy. Annual Review of 16. Cle´ment G (2006) Swenzia, n. nov., a replacement name for the preoccupied Ecology and Systematics 23: 449–480. coelacanth genus Wenzia Cle´ment, 2005. Journal of Vertebrate Paleontology 26: 44. Queiroz de K, Gauthier J (1994) Toward a phylogenetic system of biological 461. nomenclature. Trends in Ecology & Evolution 9: 27–31. 17. Lund R, Lund WL (1985) Coelacanths from the Bear Gulch Limestone 45. Sereno PC (1999) Definitions in phylogenetic taxonomy: critique and rationale. (Namurian) of Montana and the evolution of the Coelacanthiformes. Bulletin of Systematic Biology 48: 329–351. Carnegie Museum of Natural History 25: 1–74. 46. Sereno PC (2005) The logical basis of phylogenetic taxonomy. Systematic 18. Millot J, Anthony J (1958) Anatomie de Latimeria chalumnae. I. Squelette et Biology 54: 595–619. muscles. Paris: Centre National de la Recherche Scientifique. 122 p. 47. Queiroz de K, Gauthier J (1992) Phylogenetic Taxonomy. Annual Review of 19. Hensel K (1986) Morphologie et interpre´tation des canaux et canalicules Ecology and Systematics 23: 449–480. sensoriels ce´phaliques de Latimeria chalumnae Smith, 1939 (Osteichthyes, Cross- 48. ICZN (1999) International Code of Zoological Nomenclature, fourth ed. opterygii, Coelacanthiformes). Bulletin du Muse´um national d’Histoire naturelle London: International Trust for Zoological Nomenclature. Section A Zoologie, biologie et e´cologie animales 8: 379–407. 49. Sereno PC (1998) A rationale for phylogenetic definitions, with application to 20. Schaeffer B, Gregory JT (1961) Coelacanth fishes from the continental Triassic the higher-level taxonomy of Dinosauria. Neues Jahrbuch fu¨r Geologie und of the western United States. American Museum Novitates 2036: 1–18. Pala¨ontologie Abhandlungen 210: 41–83. 21. Cloutier R (1996) The primitive actinistian Miguashaia bureaui Schultze 50. Agassiz L (1844) Recherches sur les poissons fossiles [5 volumes]. Neuchatel: (Sarcopterygii). In: Schultze H-P, Cloutier R, editors. Devonian fishes and Imprimerie Petitpierre (1833–1844). 1420 p. plants of Miguasha, Que´bec, Canada. Mu¨nchen: Verlag Dr. Friedrich Pfeil. 51. Huxley TH (1880) On the application of the laws of evolution to the 227–247. arrangement of the Vertebrata and more particularly of the Mammalia. 22. Friedman M (2007) Styloichthys as the oldest coelacanth: implications for early Proceedings of the Zoological Society of London: 649–662. osteichthyan interrelationships. Journal of Systematic Palaeontology 5: 289–343. 52. Romer AS (1955) Herpetichthyes, Amphibioidei, Choanichthyes or Sarcopter- 23. Zhu M, Yu X (2002) A primitive fish close to the common ancestor of tetrapods ygii? Nature 176: 126. and lungfish. Nature 418: 767–769. 24. Hensel K, Balon EK (2001) The sensory canal systems of the living coelacanth, 53. Cope ED (1871) Contribution to the ichthyology of the Lesser Antilles. Latimeria chalumnae: A new instalment. Environmental Biology of Fishes 61: 117– Transactions of the American Philosophical Society 14: 445–483. 124. 54. Lund R, Lund WL (1984) New genera and species of coelacanths from the Bear 25. Millot J, Anthony J (1965) Anatomie de Latimeria chalumnae. II. Syste`me Nerveux Gulch Limestone (Lower Carboniferous) of Montana (U.S.A.). Geobios 17: 237– et Organes des Sens. Paris: Centre National de la Recherche Scientifique. 131 p. 244. 26. Nelson GJ (1969) Gill arches and the phylogeny of fishes, with notes on the 55. Schultze H-P, Fuchs D, Giersch S, Ifrim C, Stinnesbeck W (2010) Palaeoctopus classification of vertebrates. Bulletin of the American Museum of Natural pelagicus from the Turonian of Mexico reinterpreted as a coelacanth History 141: 475–552. (Sarcopterygian) gular plate. Palaeontology 53: 689–694. 27. Schwimmer DR (2009) Giant coelacanths as the missing planktivores in 56. Schwimmer DR (2006) Megalocoelacanthus dobiei: morphological, range and southeastern Late Cretaceous coastal seas. Geological Society of America, ecological descriptions of the youngest fossil marine coelacanth. Journal of Southeastern Meeting Abstracts 41: 6. Vertebrate Paleontology 26: 122A. 28. Wendruff AJ, Wilson MVH (2012) A fork-tailed coelacanth, Rebellatrix divaricerca, 57. Medeiros MA, Carvalho de MSS, da Silva MAM, Pereira AA, Araujo MN, et al. gen. et sp. nov. (Actinistia, Rebellatricidae, fam. nov.), from the lower Triassic of (2011) O senhor dos rios: Mawsonia gigas (Mawsoniidae) do Cenomaniano do Western Gondwana. Journal of Vertebrate Paleontology 32: 499–511. Maranha˜o, Brasil. In: Carvalho IdS, Srivastava NK, Strohschoen Jr O, Lana 29. Lambers PH (1992) On the Ichthyofauna of the Solnhofen Lithographic CC, editors. Paleontologia: Cena´rios de Vida. Intercieˆncia, Rio de Janeiro. 575– Limestone (Upper Jurassic, Germany). Groningen: Ph.D. dissertation, Rijksu- 583. niversiteit. 336 p. 58. Wenz S (1980) A propos du genre Mawsonia, Coelacanthe ge´ant du Cre´tace´ 30. Brito PM, Meunier FJ, Cle´ment G, Geffard-Kuriyama D (2010) The histological infe´rieur d’Afrique et du Bre´sil. Me´moires de la Socie´te´Ge´ologique de France structure of the calcified lung of the fossil coelacanth Axelrodichthys araripensis 139: 187–190. (Actinistia: Mawsoniidae). Palaeontology 53: 1281–1290. 59. Hennig E (1951) Trachymetopon liassicum, Ald., ein Reisen-Crossopterygier aus 31. Schaeffer B (1952) The Triassic coelacanth fish Diplurus, with observations on the Schwa¨bischem Ober-Lias. Neues Jahrbuch fu¨r Geologie und Pala¨ontologie, evolution of the Coelacanthini. Bulletin of the American Museum of Natural Abhandlungen, Stuttgart 94: 67–79. History 99: 29–78. 60. Maisey JG (1991) Mawsonia Woodward, 1907. In: Maisey JG, editor. Santana 32. Cle´ment G (1999) The Actinistian (Sarcopterygii) Piveteauia madagascariensis fossils: an illustrated atlas. Neptune City: New Jersey: T.F.H. Publications. 317– Lehman from the Lower Triassic of Northwestern Madagascar: a redescription 323. on the basis of new material. Journal of Vertebrate Paleontology 19: 234–242.

PLOS ONE | www.plosone.org 27 November 2012 | Volume 7 | Issue 11 | e49911